456. Economic Growth in the Age of Automation with Carl Benedikt Frey

Professors FM.

You're listening to the Unsiloed podcast with Greg LeBlanc, produced by University FM. Unsiloed is a series of interdisciplinary conversations that inspire new ways of thinking about our world. So wherever you are, enjoy today's episode, and here's your host, Greg LeBlanc.

Welcome to Unsiloed.

This is Greg LeBlanc.

Greg LeBlanc, and I'm here today with Carl Benedikt Frey, who is at Oxford University at the Oxford Martin School, also at the Program on the Future of Work and AI. I will get into that. Also visiting at the University of Lund in economic history. Welcome, Carl.

My pleasure. Good to be with you.

Well, I forgot to also mention that you wrote this book called The Technology Trap: Capital, Labor, and Power In the Age of Automation. And this book covers a lot of the ground that I'm interested in, which are kind of the causes and consequences of economic growth, but also kind of the obstacles to economic growth. So one of the mysteries that we're all interested in is why we saw this huge discontinuity in the 18th century, the economic revolution known as the Industrial Revolution. And I think There's a lot of stories that have to do with exogenous technology shocks and so forth, but I think you focus on some of the obstacles, some of the institutional and political obstacles. And you point out that productivity increases and economic growth, these things are never Pareto improving. There's always winners and losers. And the pace of technological change is driven in large part by the power that is wielded or not wielded by the losers and the extent to which the losers can sort of put sand in the gears. And you argue that until the late 18th century, early 19th century, these powerful interests that could block growth, they had a lot of power and that power got stripped away, particularly in England. And I guess when you bring this argument all the way to the present, one of the questions that we all have is, How are the winners and losers interacting with each other in the political realm today to either block or permit this economic growth? So I guess I wanted to start out, you have this wonderful quote, and I think it's an anonymous quote, that economists are kind of history deniers. We always focus on the present. We don't seem to think history can offer us a whole lot of insight. That happened in the past. It's a different data point. We're in the present, we got to look at things from the present perspective. And you're primarily interested in issues of contemporary economic change. Why do we need to know anything about history?

Well, first of all, history is the only thing that we have to go on, right? So we can speculate about the future, but in the end of the day, the only thing that we do know about the future is that we are pretty bad at predicting it. And so If you want to say that the future is likely to be very different from the past, then at the very least, I think we should be able to state why. So I think history should always be our starting point. And the conclusion does not need to be that history always rhymes. That doesn't have to be the case. Some things rhyme better than others. And, you know, obviously a key part of trying to figure out what's likely to happen with a big technological change like artificial intelligence, for example, is trying to look at previous technological transformation and learn from those episodes. So that was my starting point with the book that you kindly mentioned, The Technology Trap. And a key theme in that book is that we've been through quite different episodes of technological change in the past, and that people's reactions to new technology to a large degree depends on what kinds of technologies we are developing. Are we developing automation technologies, replacing technologies that are essentially doing what we are already doing, but replacing people in existing activities? Or are we developing technologies that are more likely to create new types of jobs, new types of activities, new types of industries that then absorb labor? And we shouldn't expect people to react the same way to those kind of technologies. If you take the automatic elevator, for example, all it did was getting rid of the operator. If you take the telescope, well, it allowed us to look at the moons of Jupiter and do new and previously inconceivable things, but as far as I know, it didn't replace a single person. So we shouldn't expect people to react in the same way to those different types of technologies.

Right. So whenever there's any kind of economic growth, there's going to be these losers, and their losses may be temporary. They, you know, depending on how you define temporary, right? It could last a long time. And the new jobs that are created, they might be in very different places. They may require skills that these folks lack and so forth. But you highlight how until the late 18th century, it was very, very difficult for any kind of new technology to get adopted. Why is that? Who were the primary blockers? I mean, you said that the powerful people, the elites, had to have a stake in economic progress, but the folks who were blocking it, were they the elites or were they— another group, the artisans and, and so forth?

Well, a key institution that persisted all over Europe from the 12th century onward, and that was very strong in China as well, that we can come back to that, were the guilds. And in particular, in my book, I'm focusing on the craft guilds. And craft guilds certainly aided technological change in some ways. They were a way of transmitting knowledge about production and technology from one generation to another. And to the degree that they permitted entry, they also to some extent facilitated horizontal knowledge flows, though that was more limited. But they fiercely resisted anything that threatened their own jobs and incomes. And I think that is only natural. But if you think of the first wave of technological innovations, what they sort of collectively did was create the factory system. And the factory system essentially replaced the domestic system and artisan shop. And so the jobs and incomes of these craftsmen were very much threatened by the coming of the Industrial Revolution. So it should be no surprise that people resisted it. And we all know about the Luddite riots. Well, they were only part of a long wave of riots that swept across Britain. That swept across continental Europe, that swept across China as well. And the key difference, as I point out in the book, is that the Luddites were unsuccessful in resisting technological change. In Britain, the Parliament was very much committed to siding with the innovators in this period and would not essentially undermine anything that they felt threatened Britain's position in international trade. And because Britain had relatively high wages, mechanization was absolutely essential for British goods to be competitive. And so on the one hand, Parliament made sure that petitions to block the introduction of new technologies from the 7th century onwards were essentially unheard. But they also provided a very active role in actually funding the road network through the turnpikes, which essentially meant that markets in Britain became more integrated than they were elsewhere in Europe. And certainly in China, where cities were much further apart. And because the political clout of craft guilds did not extend beyond the city in which they operated, the integration of markets essentially meant that one guild adopting modern machinery could produce more cheaply and thereby undercut other guilds elsewhere. And what that meant was essentially a race to the bottom and the end of the craft guilds in Britain. That process first accelerated there, and every other place we know of in Europe that modernized essentially ended up abolishing the guilds, whether it was forcibly through the introduction of Code Napoleon in Germany, for example, and of course, in France too, or whether it was through, as in Prussia, revolution from above under Stein-Hardenberg, reforms that abolished the craft guilds. And even in Russia, in the Russian Empire, that happened much belatedly after the losses of the Crimean War.

So where did these guilds get their power from? I mean, these were non-democratic societies. Was it just that the rulers feared Unrest, these rulers didn't hesitate to slaughter people when they got in their way, right? So what was the source of power of the guilds?

So first of all, they were well organized, right? And we know the logic of collective action, right? A few people that are very well organized, they are better to mobilize and get support. And they got a lot of support, often from monarchs and local princes. And they sometimes also played a very active role in town politics in the Dutch Republic, for example. So the source of the power somewhat varied, but in the end of the day, the power rested on being well organized and the threat of unrest and the fact that most rulers did not want to see anything that potentially undermined their political clout. If we look, for example, at the revolutionary era in France, there was nothing like the labor repression that Britain experienced during the First Industrial Revolution. And I think it's important to remember that the troops that was sent out against the Luddites. The army was actually larger than the army that Wellington took against Napoleon to the Peninsular War in 1808, right? So there was an extraordinary amount of labor repression going on in Britain during this period. And in revolutionary France, for example, there was no government that was able to mobilize in the same way that Britain did.

The guilds, they were interested in protecting their monopoly rights. Why couldn't they use their monopoly rights but also adopt the new technology? I mean, you describe in The Second Industrial Revolution in America how the unions, the unions actually played a very important role. Even though we had very violent labor conflicts in America in the late 19th, early 20th century, the unions were never interested in slowing down or destroying technology. Is that because of a change in the nature of technology, or is it just that the form of organization that we had with the unions was much more effective than the form of organization we have with these guilds?

It's a very good question. So I think obviously to some degree it is a result of a commitment problem, right? So if you can commit to redistributing income down the line and making the guilds better off through industrialization in the end of the day, then clearly you have a way of potentially alleviating the unrest that we saw. And I think it's important to remember that in the case of Britain, actually the regions where the Poor Laws were relatively generous, so less unrest as a result of rapid mechanization, right? Britain during this period was the only country on this planet that taxed itself at roughly 2 to 3% of national income to provide for the poor, and that arguably eased the transition to the modern world. In the 20th century, and obviously from the 1930s onwards, in particular after the Great Depression, the welfare state emerges. Technological change becomes less disruptive as a result of that more broadly. And we begin to have more stable labor-management relations where, you know, what you do in one year has an impact on the relationship and your ability to manage technological change and manage industrial relations more broadly, also has consequences down the line. And I think as long as the sort of mass production system was relatively stable and growth essentially depended on churning out more goods at an ever faster pace, then not having too much turnover and having above all peace on the factory floor was extremely valuable. And so it was relatively easy to negotiate settlements whereby workers could simply gain from automation in terms of better pay, shorter work weeks, more holiday, sick leave, etc. But I think it's also true that this was a period of very rapid globalization with expanding markets. And it was also a period of time where we saw a lot of enabling technologies creating vast new industrial undertakings like the automotive industry, for example, and all of its supplier industries and all of the industries that produce the machine tools to produce the components going into the automobile and the rise of road commerce and tourism, et cetera, et cetera. So I think there were a lot of new tasks being developed during this period as well, which meant that people had very good outside options. And so, you know, you had to offer people a good package in order to make sure that you could keep people in the first instance.

Well, I think a big theme of your book is you're trying to figure out whether or not the automation revolution, the computer revolution that we've been undergoing for the last couple decades is more like the first industrial revolution in England or the second industrial revolution in the US. And you point out that the key differences there is that with the first industrial revolution, there was this thing you called Engels' pause, after Friedrich Engels, where the quality of life of a typical worker didn't really go up. They didn't really participate in the increases in productivity. In fact, their living conditions may have gone down. You know, how is that possible? Because one would think that if you increase productivity, more output per worker, right, at least according to standard economic theory, those workers ought to be compensated at a higher rate. So what accounts for the Engel's pause?

It depends on the types of technologies that we are developing. So whether to what degree they are replacing or enabling, to what degree they're actually augmenting people in existing activities or whether they are replacing them. And then also the productivity effect in terms of offsetting the displacement effect and creating new demand and income elsewhere in the economy, and then how easy people can transition between those types of activities. And sort of the framework I'm borrowing from here in the book is the framework of Daron Acemoglu and Pascal Aristrepo, which some listeners might be familiar with. And so what you see during the first industrial revolution is that early spinning machinery in particular is designed to be tended by children. So one way of, you know, really undercutting the bargaining power of highly skilled craftsmen is to make these machines in a way that you can get anybody to essentially operate it, and you can undercut their bargaining power as a consequence. of that. And what we see as a gradual process, once the modern factory system takes off, is that artisan industries are essentially crushed. And those people don't find ample job opportunities in factory towns because, again, these spinning machines are specifically designed for child labor. That sort of somewhat changes when the steam age comes along. And we tend to often associate the first industrial revolution with steam power, but steam power is actually a latecomer to the industrialization process. And when steam power comes along, much of the heavy machinery becomes sufficiently heavy and complex that you actually need people with higher levels of skill, and you begin to create some new tasks. and activities on the factory floor. But for a very long time, you have this process of labor-replacing technological change that drives down wages for ordinary people, even as the British economy takes off. And if you look at the aggregate data from 1760 up until 1840, the British economy grows by something like 50%. Wages on balance increase are roughly 12%, but they're probably falling at the lower end of the income distribution, right? So if you look at other indicators of wellbeing, such as consumption or people's heights, nutrition, and so on, they tell a similar story. And so this is definitely not a period where prosperity from technological changes widely shared.

Well, you talk about a lot about pushing and pulling into the new workforce. So how much of this was pulling and how much pushing? I mean, were these employees— I mean, you said they're mostly child labor, but were the employees being pulled in by new opportunities or were they more or less forced into this new employment by the destruction of old employment opportunities?

I think the push factor here is very significant, right? And We should not romanticize or over-romanticize the domestic system, right? But it was a time when people worked from home surrounded by their family and children, and people worked at their own pace, not at the pace of the factory clock.

It was the original work from home.

Yes, exactly. It was basically what we're doing now. But if a harvest was bad, then you had a pretty rough year, right? put it mildly. So it was not quite, you know, what we're used to these days. But nevertheless, people have a certain degree of autonomy. And in factory towns where people then began to cluster, obviously you have this bad sewage system, spread of disease, et cetera, et cetera. And so life expectancy is many, many years lower in places like Manchester around this time relative to the countryside.

Now, we don't talk about the Engels pause in the US. I mean, is that simply because the value of labor was so high that the frontier was there so that textile mills in Waltham, let's say in Massachusetts, they weren't creating these horrible labor conditions because the outside option was just too good?

Well, I'm not sure. I mean, we're not calling it Engels pause, but I do think that, you know, go back to the 1980s and if you look at the trajectory of wages and if you go to, you know, old steel towns and cities, you can see and hear what has been going on there over the past couple of decades.

No, I meant in the early 19th century in the US.

Oh, I see.

We don't talk about the horrible labor conditions. You know, we don't have our Charles Dickens going to the factories in Massachusetts in the 19th Yeah, no, that's right.

And so in the United States, I think it's more, somewhat more ambiguous. So you have child labor in the United States as well as a result of mechanization. But if you look at other latecomers, so the United States is relatively close to the frontier from the beginning. And so is Prussian Germany, by the way, but even in Germany, you have much more developed educational system from the very beginning. And also you don't really have child labor to the same degree because the educational system essentially developed before the rise of modern industry, right? So Engels' pause is more of a Anglo-Saxon phenomenon than it is a global phenomenon, I would say, and it's primarily an English phenomenon. And it's something we see more in frontier economies, generally speaking, than among latecomers who often catch up using a different set of institutions. And if you look at the impacts of the computer revolution today, its impact on inequality has been much more pronounced in the United States, which is the leader in computer technology relative to Europe, for example.

So why was the Second Industrial Revolution different? You see a lot of similarities. I mean, if you look at the type of work on the factory floor in Ford's factory, a lot of this work was not terribly skilled. I mean, part of the goal of the you know, Remington and Colt factories with interchangeable parts was to reduce the job of the factory worker to something that required relatively little education. So why was that movement not equally bad in terms of consequences for labor?

So if you think of the first industrial revolution, or at least its early stages, is essentially the mechanization of cotton. And later on you have the railroads and you have the rise of chemical industries and so on. And within the 2nd Industrial Revolution, you essentially have electrification, the internal combustion engine, and the continuation of what's happening in chemicals and pharmaceuticals, et cetera, right? And what you're seeing with that is a host of new industries being created. With the 1st Industrial Revolution, it's essentially the mechanization of existing activities and the cheapening of textiles. And so basically everything you can think of that you have in your home is an artifact of the Second Industrial Revolution. From your iron to your oven to your dishwasher, all of these gadgets created new types of industries, new types of tasks and activities for labor to do. And we mentioned the automobile earlier, which became the largest industry in the United States by 1940, right? And that's not even considering all of the supplier industries Et cetera, right? At this period, you have a lot of new activities being created for labor, and we don't have robots yet, right? So you need people to operate machinery. And at the same time, you know, we have a reorganization of the factory that gets rid of shafts and countershafts and, you know, You can sequence the machines according to sort of natural flow of production, which gives rise to mass production. And you need labor at every given activity in that production process, right? So there's a host of things coming together. I think the predominant difference is that just a lot of new types of activities being created that didn't exist before. Whereas during the first industrial revolution, primarily centered on automating things that we already do.

Were people less likely to resist technological change because they were aware of other opportunities? You know, if you're a farm worker and now we have this new reaper that comes in that effectively makes you redundant, are you less likely to protest because you know, well, okay, I can, all I need to do is kind of move to the city and I can do something else that is more lucrative? Because there it's not actually helping me to be a better farm worker, right? I'm basically out of a job. But is it really that there are opportunities that these folks are aware of that are within their grasp that leads them to resist less than they might be expected to?

So to be clear, I'm not suggesting that there was absolutely no resistance during this period. So I talk about, and I begin the book talking about the lamplighters who were certainly victims of electrification, right? So before the arrival of electric streetlights, every city used to have a team of lamplighters or many teams of lamplighters who carried torches and ladders and ignited street lamps, oil lamps at night, right? And when electrification happened, well, their skill was no longer particularly valuable.

Yeah.

Even before, you know, electric streetlights were controlled from substations, every lamppost had its own switch, which was very straightforward to flick on and off. And so easy that kids can do it on the way to school and back home from school, right? And so lamplighters did go out and protest against electrification. And we see Similar things happening with longshoremen and switchboard operators going on strike as well. So these things happen, right? But in the labor market where you have more outside options and where you have stronger pull factors, these episodes were clearly relatively few. And plus, people during that time had at least experience of economic growth and what improved living standards looked like. People didn't really have much of a conception of that if you go back to the 19th century either. And so in the 20th century, technology becomes something that people begin to believe in and something that they can actually improve standards of living. In the 18th century, I mean, it wasn't the iron law of wages doesn't come from Marx, right? It was something that Ricardo and Malthus and essentially everyone believed in, because there had been very little improvements in people's living standards over the preceding centuries. Not to say that there weren't any improvements, that's wrong. There were improvements, but they were quite modest. I think for people in the 20th century, just the collective experience of the progress that people were seeing around them, And the greater outside options that people faced, and plus the welfare state, led to an environment where resistance against technological progress was no longer the norm. But that is not to say that there weren't any resistance. And actually, most resistance in that period coincides with economic downturns, right? So the Great Depression, the 1930s. number of recessions during the 1960s where all of a sudden we all know it's worse losing your job in the middle of recession than during a boom, right? It's true if your job is automated as well.

Well, I want to dig into this distinction. This is a distinction that's often used in the field between the labor-replacing technology and labor-enabling technology. And I think sometimes when people think about these enabling technologies, they think that it's all happening within a particular profession, right? So, you know, there's this famous retort that people have about bank tellers, and they say, oh yeah, you know, the ATM, it didn't eliminate bank tellers, it just changed their job, right? And we have more bank tellers than we did before the ATM. I mean, I think that's old data. I can't imagine that that's still true. I think They're on their way out, right? But people would point to an example like that where they'll say, oh, look at Excel. Excel didn't eliminate accountants and financial analysts, it just created more demand for them. But for jobs like truck drivers, you talk in the book about how that's the number one profession in about half of the states. These folks are going to be gone. I mean, there's no way, even if we have remote truck driving, there'll be One truck driver for every 10 trucks, right? And so presumably this technology is wiping out that entire occupation, but the spillover effects will lead to new jobs in other areas, jobs that no one could have envisaged 20 years ago, right? Uber driver was not an occupation, right? You know, lamp lighting didn't exist for very long. I mean, they didn't have gas lamps, and then they had lamp lighters, and then the lamp lighters disappeared. You know, when we think about what happens to these folks, sometimes they reinvent themselves within a profession, sometimes they have to transition to a completely different profession altogether. Are both of those examples of enabling technologies because they maybe create more opportunities than they destroy?

First of all, there are cases that are clear-cut replacing or enabling. So I mentioned the telescope. And the elevator. But then if you take a technology like the automobile, right? I mean, it did replace the horse carriage, right? So—

If the horses could vote, they might not have been in favor of cars, right?

Well, exactly. I think Leontief was right on that one, right? If they could have voted, they would've voted against it.

Well, I don't know. The average life of a horse today is pretty darn comfortable. You know, I have horses and, you know, they only work a couple hours a week, you know? So it's a pretty cush life nowadays to be a horse. There's just a lot fewer of them.

That might be the case, but to stick with humans, there are a lot of technologies which have some replacing components to them and some enabling components to them. And essentially, you know, it becomes an empirical question whether these technologies have created more tasks and activities. for labor than they have replaced. And I think that's true for a pretty wide range of technologies. And, you know, we have pretty compelling empirical evidence now that during the sort of middle part of the 20th century, there was a wide range of new tasks being created at the middle of the income distribution. If we sort of fast forward to the 1980s, you begin to see that more and more new job titles are related with computer technologies in particular, but they're also to some degree created with new consumer demands, right? So sommelier is a job title that emerges for the first time in the 2000s, right? And that's not a result of technological change, but it might be a result of people in Silicon Valley enjoying wine tastings, for example. Right? And so there is more complexity there in the sense that there can be very different tasks being created by enabling technologies. And some of that may be very easy to move into, and some of those might be relatively hard to move into. But if we just look at the aggregate, there's clearly been a shift towards less enabling, less job-creating technological change. And more automation. So I think that's safe to say, but as you did, we can add more nuance. And to further build on what you said, right? So if you take the automatic elevator, for example, it did replace elevator operators, but in some places like the Chrysler Building in New York City, for example, some of those were just shifted to become porters or other activities within those buildings. So, you know, that was a replacing technology, but it wasn't necessarily very bad. Whereas, for example, the bank teller, well, if you're a bank teller, you can only tell a currency, right? So you're completely useless in a modern bank, right? So the fact that that occupation changed doesn't necessarily mean that all of those bank tellers were better off as a consequence. of those changes. And in similar fashion, some people tell me sometimes that when autonomous vehicles and autonomous buses arrive, we have a lot of school buses. So, you know, you will have to have somebody on the bus to look after the kids. So even if the bus drives itself, you need somebody at the bus, right? But that person is not somebody with a driver's license, it's somebody who's good at looking after kids. It's a completely Different person, different skill set, right? So there is additional complexity there, and I think it's important to be clear about that and unpack that. But it's also clear that, you know, these categories, replacing and enabling, do tell us something meaningful about how widely income in society is shared and the kind of outside options that people also have during labor-replacing technological change. If you have more enabling technologies to counterbalance that, it makes life easier for you.

Is the distinction between a labor-replacing technology and a labor-enabling technology a function entirely of the technology, or is it also a function of the skill base? You could imagine that when you adopt a new technology, it might create demand for new types of work, but the folks who are displaced don't have those capabilities. And so it's simply replacing. But if they had those capabilities and they could somehow redeploy their human capital into those new areas, then it would be enabling, right? So are those 2 characteristics, the skill base or the education base or the redeployability of the human capital, And the nature of the technology, I mean, can they be discussed independently of each other when evaluating the extent to which the new technology is replacing or enabling?

Well, to some degree, I think it can, right? So elevator operators are, you know, as far as I know, not a feature of modern economies in the United States, Sweden, Denmark, Italy, Spain, et cetera, et cetera. But as I mentioned earlier, I'm the first to admit that technology does interact with institutions in meaningful ways. And as I mentioned, you don't see an Engels pause in Germany, for example, partly because you have educational institutions that really ease the transition. And so that clearly helps, but I don't think that is the full explanation of what's going on. If you look at what's happened to inequality post-war years versus post-1980, something meaningful has happened across the OECD when you compare those 2 episodes. And we can discuss relevance of different factors there, but there's also a lot of variation between places in the post-'80s era, right? So the rise in inequality has been larger in the US than it has been in Denmark, for example. That's absolutely clear. But inequalities have been increasing in Denmark too.

Well, I was thinking about this race that Tim Bergen talks about between education and technology, and it seems like at least in the US, the education was advancing quickly enough that it could keep up with the technology. Is the problem today that education's not keeping up, that the workers who are displaced by these technologies simply don't have the skills that would be needed for them to find alternative employment in the areas which are now in high demand?

I think it's part of the story, right? But the race between technology and education is a world in which everybody's better off, right? And that has not been the case. So we need to somehow modify That model of the world. And what we've seen since the 1980s, in particular across advanced economies, but also in some emerging economies, is labor market polarization and the decline of middle-income jobs, right? And so the race between technology and education view of technological change does not explain that part of the story, right? And that's where sort of the task-based view and things like replacing versus enabling technologies do have some explanatory power.

You talk about, for instance, the geographic separation between the areas where labor-replacing technology has really taken off. Those are kind of the areas where you see lower labor participation rates, you see living standards stagnating and so forth. And then the other areas of the country where it's labor-enhancing technology, and this is really where There's higher labor participation rates and so forth. So is part of this not just a mobility issue across occupations, but also kind of a mobility issue across geographies?

I think the geographic component does play a very important role in the United States. At the very least, we do know that mobility across states has been steady decline since the 1970s. Or so, and so the fact that people are not moving towards places where better opportunities exist, I think is a real challenge. And you know, I think there are a variety of explanation for that, including the cost of housing, the cost of moving, people being more tied to their communities. Some people. More on the right tend to say it has more to do with benefits and the fact that you no longer have to move. And, you know, it's certainly true that if you go back a century, everybody had to work. And so labor force participation among prime-age men was very close to 100%. So there is a variety of factors which have contributed to lower labor mobility, and might also be things like occupational licensing and the fact that, you know, it's not harmonized across states and space and etc. My wife is a medical doctor. If we were to move to the US, she couldn't work, right? It's not increasing the likelihood that we're going to move to the US at some point. These things do matter. But again, if you go back to the 1970s and '50s and '60s, and you look at the new types of work being created, it was relatively routine, semi-skilled type of work, right? So The type of work that has been created since the 1980s is fundamentally different in that it's more related to higher levels of education, cognitive type of skills, right? And those new jobs have primarily been created in cities with an abundance of skill, right? And so if something is knowledge intensive, knowledge spillovers matter more, your ability to learn from each other and the proximity to other knowledge workers. Whereas if something is more routine and standardized, those knowledge spillovers matter less. So we have more of a clustering of those particular types of jobs. And we, in addition to that, have a situation where, I mentioned some layers earlier, I don't know too many people that do virtual wine tastings. Maybe they exist now, but for most of these in-person service type of jobs, they need to be done where people that demand these services live and work. So we're creating a situation in which people doing in-person type of services also preferably needs to live where these new tech jobs are being created, whereas we're seeing many of these routine rule-based activities being automated away. So I think it's a combination of these changes that we're seeing in technology, and it doesn't make it easier that people seem less willing to move than they did 4 decades ago.

Yeah, I mean, you talk about AI, right? And the impact AI is having. And of course, AI is just a continuation of this computer revolution that we've experienced. But you also talk about how machine learning is very different from smart systems. And the book was written in 2019, so you didn't even get to talk about generative AI. But each of these new generations of AI, they seem to have different types of impact on the workforce, perhaps moving more from enabling to replacing, right? Should we be thinking of this new revolution as being more like the first than the second industrial revolution?

I think it is more like the first industrial revolution, and I still think that. And I can't think of a single AI application which is not about automation or doing something that people are already doing a bit more productively, whether it's writing or coding or image generation. Even AlphaFold, right, which is sort of the poster child for AI in science, is folding proteins. And, you know, I don't want to diminish that accomplishment in any way. It's a tremendously useful tool if you speak to people in the field. But people were folding proteins before then at a much slower pace. been a huge help, but every application I can think of is in the end about automation. And so I do think it's more like the first industrial revolution than the second one, so far at least.

But do you agree it's going to dramatically increase productivity? I just gave a talk on Solow's famous quote that computer age is everywhere except in the productivity statistics. And a lot of people, including Erik Brynjolfsson, have kind of followed up on Paul David's idea that this is sort of an artifact of the way in which we are measuring productivity and that ultimately there will be no doubt that the computer revolution will increase productivity, right?

Well, the computer revolution did increase productivity, but only for about a decade, it seems like. And so the question is, you know, I have no doubt in my mind whether the AI revolution or whatever you want to call it will increase productivity. The question is by how much and for how long a period of time. And obviously automation at the end of the day is a one-off, right? So if all that AI is about is automation, well, then the future of productivity simply depends on the potential scope of automation, so to speak. And that will then eventually Peter Roth, whereas if it's about creating new tasks and new products and new innovations, then it can be more sustained, right? And I think that's a key reason that the Second Industrial Revolution lasted for a very long period of time because it created a host of entirely new types of economic activities. And so I think a key question going forward is. Can we design our institutions or, you know, help make sure that AI is more being used to create new activities? I think it is likely to have a much more sustained impact on productivity growth going forward.

But I mean, aren't these new forms of machine learning and artificial intelligence, generative AI, I mean, they're primarily replacing the work of the symbolic analysts, right? I mean, the artisans of our time, the accountants and the lawyers and the doctors and so forth. But it isn't really doing much for our locksmiths and our waiters and, you know, the people who are engaged in more manual tasks, at least for now. Shouldn't that lead to another form of compression? Most of us would think that these new tools would just make education more and more important, right? And we would become involved in more and more abstract exercises as we become software engineers. But it's not clear that that's the case now. I mean, we're not going to need all these software engineers. What we're going to need are we're going to need more bricklayers, we're going to need more waiters, we're going to need more locksmiths. I mean, what's astonishing to me is that here in the Bay Area, Locksmiths, roofers, landscape architects, all of these people are coming from outside of the Bay Area and they're all coming from Brazil. They're coming from Guatemala. They're coming from everywhere but other parts of the United States. And, you know, I have a friend who, she flew in a roofer from Virginia to basically do her roof for a week. And it cost about half as much to fly in a roofer from Virginia and fly them back to Virginia than it cost to hire a local Guatemalan immigrant. to do the same job. So it's in terms of the labor demand, I mean, are we seeing a complete reshuffling of the labor demand away from these symbolic engineers and towards folks who are doing stuff that's much more difficult to automate?

So I think we don't know yet. I fully agree on the manual work. So AI is not going to do much in that domain, although I think there are some instances where, you know, it will make it easier for us to fix things ourselves, right? So if something is broken, you can take a picture of it and, you know, ask, what do I need to do to fix that? Then we will have AI that can provide you with some instructions. So I wouldn't say that manual work is completely unaffected by it. You know, I'm an academic, so you can imagine how skilled I am. at these things. So even if the AI tells me what to do, it's going to fail on the execution. But I take your point generally that it's not what we necessarily thought it would like look like 10 years ago. When it comes to more, we're saying we don't, not going to need as many lawyers and software engineers and et cetera. So I think what AI, generative AI that we have now, Thus, it's essentially reducing barriers to entry into content creating professions, right? So it's a little bit like Uber, what it did for taxi services, right? You all of a sudden, you know, knowing the street of the name of every street in San Francisco was no longer a particularly valuable skill. And so with GPS technology and then the Uber app that, you know, matched supply and demand, anybody can just get into their car and earn a little bit of money on the side. So the result was more competition. In tax services, then a key question is how is demand likely to respond? And when it comes to content creation, I suspect that it's not going to respond that much. So because, you know, I'm not going to watch twice as much Netflix just because the subscription rate goes down, the context becomes a bit better, right? I just don't have the time. I mean, you can argue that AI could have some impacts on the demand side too. So you can just feed a number of reports to your AI algorithm and say, sort of summarize this for me. And so you can sort of process more material as well, but I suspect that's not going to change demand radically. But I also don't think that the LLMs that we are using today is like the future of AI. I don't see them being, you know, the technology that is sort of fundamentally going to transform our economies. I think they are extremely useful for many things. They are Very impressive. I don't want to downplay it, but I don't think we are going to have a sufficient amount of good training data going forward to see the improvements that we've been seeing over the past couple of years. I don't think that synthetic data is very likely to make up for that for reasons we can go into. I think there are questions over the continuation of Moore's Law, but more fundamentally, it doesn't seem to me that these algorithms learn generally new capabilities just by scaling up existing models. And in fact, there are a number of studies where, you know, you feed the AI a set of questions it will have been trained on, And a set of questions that would only have emerged after the cutoff date up until which it has been trained. And it performs extremely poorly on the latter, which suggests, you know, it's doing memorization rather than generalization. So I think, you know, we will need quite a bit of additional innovation to really get sustained productivity growth. If progress in AI stops today and we could just harvest the returns from what exists. I think it will still be a significant one-off in certain domains, but it's not something that would measure up to the Second Industrial Revolution, and it would probably not quite measure up to the computer revolution either.

Well, look, it's difficult to understand contemporary American politics without also acknowledging that there have been losers folks that have seen their prospects and expectations of prosperity limited by technological advances. But it's also— it seems like it really hasn't done much. That discontent hasn't really done much to slow down technological progress. Do you think that if technological progress starts to impact adversely more elite professions, that it could potentially lead to some actual slowdown in technological progress, or is the cow out of the barn to the point where there is no interest that could organize sufficiently well to slow down progress in this area?

So I think it's certainly true that few of the factory job workers who lost their jobs to robots in the 1990s wrote op-eds for the New York Times or the Financial Times. And it's also true that there were a few people calling for a robot pause during that period, right? So I certainly do think that when people with more political clout see their incomes being impacted, it's more likely that you will see some resistance. But I think the hurdles to technological change are actually broader than that, right? So If you look at the sort of 10 most valuable companies in the United States, in Germany, in France, in Italy, in Britain, in Japan, right? The United States is an outlier in the sense that most of them are actually relatively new. In Germany, they are basically spinoffs of the Second Industrial Revolution. In France, they basically produce luxury goods. In Japan, they're essentially The caretsi who drove the post-war miracle, et cetera, right? So it seems to me that technological change is not something we should take for granted necessarily, even if the technology isn't replacing, right? So there are vested interests and special interests in most societies that are tied up in existing economic activities and are not very interested in new economic activities. And I think in the United States with the federal system, right? I mean, the Boston area used to essentially, or to be very much just the forefront of technology. The computer revolution happened in the Bay Area. So often new types of activities tend to emerge in other places that create more space for that sort of experimentation. So I don't think it's only about labor resistance. And in fact, you can easily make the case that what we've been seeing over the past couple of decades has been diminishing labor power, most certainly rising corporate power. And maybe we see a rebalancing there. But right now there is a lot of interest in existing activities coming from corporations themselves.

You mentioned there are a couple of things we could do to potentially make technological change more Pareto improving, some policy recommendations. Some of them involve training, retraining, education, but some of them also have to do with making it easier for people to move to new jobs and to new professions. What sorts of policies should we be thinking about? And a lot of these policies that you discuss, I mean, they seem to be policies that you find in European countries, but one would think that if the European countries are sort of better at making technology Pareto improving, that you'd see them doing more technology. So how do we resolve this paradox?

Yeah, so clearly, as you say, right, there's something more going on. So the Europeans do have better policies in a way for easing the adjustment of people to technological change. And as you point out, most of these technologies have come from the United States. So clearly there are some institutions or endowments that the United States has that tends to favor technological progress versus Europe.

In spite of the fact that the losers may wind up suffering more in the US.

Yes. In spite of the fact that I think that's right. So I think if you look at the US story, right? I mean, it's not just the US story. It's very much a Bay Area story for the past 3 decades or so. And there are a number of things that seemingly need to come together to develop these technologies to begin with. In Germany, in the early stages of the computer revolution, you know, people like Alfred Chandler writing in the 1970s thought the computer revolution is just, you know, like the second industrial revolution in the sense that it's the same companies that are going to drive it, right? And in Germany, it was. BASF and Mannesmann and Siemens and these companies that were at the forefront of the computer revolution in Europe, right? But none of them we associate with computers today, right? And most of the companies that we associate with mainframes, well, we've forgotten about those a long time ago, right? And the companies that were leading in personal computers didn't become the leading companies in software production, right? So when it comes to the development of these technologies to begin with, it's not an automatic process either, and it doesn't tend to happen within incumbent firms. Kodak is obviously the classic example of this, right? Most companies don't want to develop something that tends to undercut the revenue model. And, you know, There might be a reason also why Google has been slow to respond to generative AI despite developing the transformer internally, right? So I think when it comes to the innovation process, business dynamism is really important. And the United States has historically been more dynamic in that regard. Maybe that's changing. So, you know, there are A number of studies that show that barriers to entry in the United States have increased and business dynamics have fallen. So maybe it's an end of era and, you know, we can discuss that. I'm not sure it is, but in Europe, I think many of the institutions that you have in Germany and France, et cetera, was built around catch-up growth, right? They were not built for innovation necessarily, and those institutions have persisted. So I think that helps explain, you know, why the development of these technologies has tended to come from the United States rather than from, let's say, Europe. And what I do think still very much matters though, is that when it comes to the adoption, solving the collective action problem is hard, but I think There's been a very clear shift in the attitudes that people have to technology, right? Only a decade or two ago, I mean, basically every party, every person you spoke to would agree, broadly speaking, that technological change is something good, right? Recent Pew Research survey, which I cited back in 2019 when I wrote the book, suggested there would be a majority in the United States People, you know, if there was a vote on how many machines that businesses ought to be able to implement. I think more striking more recently is that people are just much, much more concerned about artificial intelligence in generally than they are about its potential benefits, right? And I think that doesn't just have to do with concerns over automation. There's just a lot of other things. tied up to it. There is, I think, a popular sense or a relatively widespread sense that technological change is not yielding the widespread benefits that it used to. And I think that is something that more and more people are sort of gradually beginning to tap into. And right now it's very much focused on big tech because I think there are some good reasons for that, but about the competition policy and so on. But it's also sort of become the face of what's wrong with society in a sense. And I wouldn't read into what you just said in the sense that in Europe we have these institutions that protect labor. and would have less innovation. And in the United States, it's the opposite. It's not clear to me that that will persist.

Well, look, I mean, I think people's attitude toward technological change depends on their beliefs about whether it will make their lives better. And those beliefs may or may not be in alignment with the reality on the ground. Are people generally good at evaluating their prospects post-technological change, or are there things that can make people either more or less anxious than they ought to be? I mean, nobody likes change. Nobody likes to have to lose their job and find a new job. But are people sometimes maybe irrationally pessimistic about their prospects, or do people generally have a pretty accurate view of, of what their prospects are?

It's a good question. So I think overall, very few have a good idea of what artificial intelligence is going to do to society, or even the people who make a living speaking about it don't have a very good idea of what the next 10 years are going to look like. At the same time, I think the actors and the screenwriters who sat down and negotiated a deal over the use of generative AI in Hollywood did have a pretty good idea about what that might do to their relatively narrow domain. So I think when something becomes uncomfortably close to what you do, when something becomes uncomfortably apparent, then people all of a sudden do jump up and say, well, I'm quite likely to be impacted by this. But that doesn't mean that there are going to be 1,000 unforeseen consequences, which are probably going to make me sound very silly if somebody listens to this podcast 5 or 10 years from now.

Well, Carl, thanks so much for joining me. I think you're absolutely right that it It makes no sense for economists to be history deniers, right? Because it's really by looking at history that we can make better forecasts, I think, about the future. The book is called Technology Trap: Capital, Labor, and Power in the Age of Automation. Thanks so much for joining me.

My pleasure.

5-star rating and review in your favorite listening app. To listen to our other episodes, please visit our website at www.unsilodpodcast.com.