Johanna looks at whether the past can give us clues about the future with Carl Benedikt Frey, Dieter Schwarz Associate Professor of AI & Work at the Oxford Internet Institute and writer of How Progress Ends: Technology, Innovation, and the Fate of Nations. They talk about whether progress will end in the context of technology, innovation and (of course) artificial intelligence. They also discuss past events like the industrial revolution and US Gilded Age can tell us about the current technological direction humanity is taking. In keeping with Tech Mirror’s deep dive into Tech and Geopolitics, Johanna and Carl also discuss who will succeed in the rise of AI – the US or China? And Carl answers the big question: is progress ending in the US?
Links:
Carl’s book – How Progress Ends: https://press.princeton.edu/books/hardcover/9780691233079/how-progress-ends
For transcript and full show notes visit techpolicy.au/podcast
Transcript
Joanna: The Tech Policy Design Institute acknowledges and pays our respects to all First Nations people. We recognize and celebrate that among many things, indigenous people were Australia’s first tech innovators.
Carl: I don’t think that the United States is in any way doomed. It can reverse trajectory as it has done many times, and it’s generally been a bad bet to bet against United States. United States has been the most dynamic economy for Century. And a half. But it’s also important to remember that, you know, nothing lasts forever.
And it’s striking that historically, for the most part, technological leadership has been a relatively short-lived experience. I live in Britain, it used to be a technology leader. It’s, you know, hard to believe today, but that was the case not that long ago, so things can change relatively quickly.
Joanna: Welcome to this episode of Tech Mirror. I’m Johanna Weaver, your host, and this is the podcast where we talk about how technology is shaping our world, but how humans can shape technology back. And today we have an absolute treat of a guest joining us from the UK, Carl Benedikt Frey, and Carl’s here to talk to us about his book, which I just love, called How Progress Ends: Technology, Innovation, and the Fate of Nations, which fits very well into the conversation we’ve been having over the last couple of episodes.
So Carl, thank you so much for joining us on the podcast.
Carl: Pleasure to be with you.
Joanna: Now Carl’s joining us from Oxford, where he is the Dieter Schwarz Associate Professor of AI & Work at the Oxford Internet Institute. And Carl holds a number of other affiliations at different parts of Oxford, but also Lund University.
And I think one of the things I find most interesting about your biography, Carl, is that you established the Oxford future of work program. So I’m quite confident that we have a lot to talk about over the next 40 minutes or so. But why don’t we start at the obvious place, which is your book title is a statement, how Progress ends, and I’d like to turn it into a question.
How do you see progress ending when we are looking at it in the context of technology and innovation? And I’m going to do the impossible thing and ask you to summarize it in a few sentences and then we’ll unpack it throughout the episode.
Carl: Well, the key theme of the book is that progress requires different people exploring different technological trajectories, and so when decision making becomes highly centralized, whether in a government or through a few incumbents that are stifling competition and entry.
That is generally speaking how progress adds, and it’s also one of the key reasons that growth was slow for so long in human history.
Joanna: And I think when people look back on that and say growth being so slow for so long, it almost sounds lovely given how fast it feels like we’re progressing now, and there’s this sense of inevitability about the way that technology is evolving.
So one of the things that you are really clear about is that progress is not inevitable. So why do you think there is? Such a strong narrative that progress is inevitable, that technology change is coming, and that we kind of have to get outta the way and respond to it.
Carl: Well, if you take my grandmother’s generation, she lived through transition from horseback to cars, the electrification of the home, indoor plumbing, air conditioning, the rise of the digital era where the computer and the internet, the smartphone, social media.
And so if you look back at that life and the vast amount of. Progress or change that happened during that period of time? I think to many people, that almost feels inevitable. She didn’t opt for her grandchildren to sit around the dining table looking at their phones. That was not her decision. So I think there, for many people, there is a sense of inevitability about it for those reasons.
Joanna: Hmm. And you use many different case studies over the course of history to emphasize or to underscore the points that technological change, that the impact that geography, culture, that institutions have. How do you see the different roles that geography, culture, and institutions have on technological progress and maybe you can use a couple of case studies to sort of start drawing this out.
’cause your book is, it’s really quite epic in terms of the breadth of the issues and the case studies that you use. It’s a fascinating gallop through history as well as a cautionary tale for us looking to deal with the circumstances we have now.
Carl: So if you take the Soviet Union, for example, most people think of it as a failure today, and in many ways it clearly was, but it also grew very rapidly over the next extended period of time, around 6% or so, over four decades.
And in fact, in the 1950s, around the time of the launch of Sputnik, many economists and respectable ones predicted that the Soviet Union would eventually overtake the United States, many pundits believed that they had found a superior recipe for progress, and then a couple of decades later of growth picked off and the Soviet Union ends up collapsing.
And so if you want to understand broad patterns of growth. You have to have a theory that explains both the growth spot and the decline. And so geography, needless to say, can’t fully do that because geography is constant. It can’t explain reversals of fortune and the geography of the Soviet Union was the same in the 1950s and just before it collapsed.
You might say that, you know, culture played a profound role and probably did to some degree, but if anything, you know, Soviet culture became more open to scientific inquiry and to discussion. And debate because no amount of barbed wire could prevent the radiant television waves from Western Europe traveling right through the iron curtains.
The sort of third bucket of explanations that sort of is institutions and things like secure private property. Well, the Soviet Union didn’t have that in the fifties or before it collapsed. And so what I argue in the book is that. If we want to understand both the growth spurt and the reversal, we need to understand how institutions and technology interact at different stages of development.
And so if you are behind the frontiers of innovation, well, you don’t have to do a lot of innovation. You can adopt technology. Invented elsewhere, and that is mostly what the Soviets did. They took advantage of the Ford Motor Company’s open door policy. They had several delegations that went there, visited, collected information, saw firsthand how the automotive industry was built, and then they built their own industry, car industry, or vehicle industry more broadly in the Soviet Union and in the era of mass production.
The Soviet system worked quite well because you could benchmark. Factory performance across plants, and so the Soviet authorities could potentially punish managers that didn’t perform that way, but that became much harder at the frontiers. Of technology because with a computer, which was meant to be this, you know, wonderful technology for planning and that would enormously help the Soviet and central planning model.
It’s actually turned out to do the opposite because when the technology is new and what you benchmark against, so monitoring and holding people accountable became harder in the Soviet system. And at the same time, the potential for inventors to explore new ideas was very. Constrained. So if you were an aircraft engineer in the Soviet Union, for example, you could go to the Red Army in Oscar funding.
If they declined, maybe had two or three different options. If they declined, well, your idea would die with you. That’s very different from the American system of decentralized finance, where Bessima Venture, famously the client invest in Google back in 1999. They probably regretted today, but it also illustrates.
The fact that Google wasn’t a safe bet at the time, right? Alta Vista and Yahoo, they were dominating. And so somebody needed to take the risk to invest in order to know if Google would catch on. And in more decentralized system, you can have more people exploring different technological trajectories. The fact that Besser.
Didn’t invest. It didn’t mean that Google was doomed. And so that is the advantage at the technological frontier. And so you can have more centralized system structure for catch up, but then when you reach the frontier, you need to switch to towards more decentralized competitive system. And because you constantly need to move between these different modes, that’s.
Key reason why many places fail to make these institutional adjustments that are so essential to progress over the long run.
Joanna: Listening to you there reminds me of one of my favorite, and perhaps is now one of my all-time favorite quotes, which comes from your book, which is that you’re saying that innovation demands breaking rules.
But efficient execution requires following them. And I think every startup can understand that startup scale up, that sort of challenge between innovation and then delivery. I think every government in the world at the moment, grappling with how to respond to AI is grappling with the how do we do that systemically, but also how do we do it in an innovative way that responds to the pace of change.
So we’ll come to how this impacts on concepts like artificial intelligence in a moment. Before we get there, I just wanna finish this part about institutional structures. Do you think it is possible to design institutional structures that are flexible enough to respond to the different points in a technology’s lifecycle?
And is there an example from history where we have seen institutions that have had that flexibility?
Carl: Well, I think for the most part you need to develop those institutions as you go along, but then once you have some of them, you can reuse them at various points in time. Right? So if you go back to the late 19th century in the United States, the gilded Age, that in large part emerged because the American state was very weak.
It was built on weak government. And, um, as a result of that, the rollout of the railroads, uh, happened through public private partnerships, and that’s created a large railroad network, but it also created federal ground for cronies whereby you had railroad companies emerging that were larger than the federal government, and you had congressmen essentially supporting this system on behalf of these railroad companies, and so what you needed was meritocratic civil service capable of regulating these giant enterprises that eventually came with a petton act, which introduced civil service examinations in the United States. Before that, you basically had a patronage system whereby. People essentially were promised positions in office in return for their political support.
And then with that civil service, you begin to see new institutions emerging, like the Sherman Antitrust Act and for example. Which eventually becomes an important tool for enforcing competition. And it becomes important in the late 20th century as well with a lawsuit against IBM, which sort of forces it to unbundle hardware and software, which opens up the door for Microsoft and other software companies to the market, and the breakup of at and t, which then becomes fundamental to the internet era.
And so. You have these turning points, so to speak, where there is a sense that something needs to be done, those institutions are created in response to those circumstances, and then you have those in place. And so antitrust, for example, I think is, is an institution that is particularly important that the frontiers of technology and not so much during the catch up phase.
Joanna: Does it worry you? Looking at what’s happening in the US at the moment that you know, I mean, there would be many people listening to this who’d say, we’re entering a period that is mirroring a lot of the cronyism that was in place during the railroad era. There is concern about the lack of regulatory capacity and the erosion of the US state.
Again, I’m aware you’re not a political commentator on today, but are there lessons that we can draw from history that maybe help us to be able to respond to what is currently happening in the US at the moment?
Carl: So I think there are good reasons to want government to be leaner and run its operations more efficiently.
But you know, the reason that, for example, the IRS is a huge operation is that the American tax code is extraordinarily complex, right? And so if you reduce staff at the IRS without making the. Tax code easier to navigate, just reducing the capacity of the government to collect taxes, right? And so if you want smaller government, the first step is to deregulate or simplify these rules and and regulations.
Otherwise, it’s just going to have the opposite effect of what you want. And then you have a tendency towards more political appointments, which is essentially going back to the. Pre Peleton Act era. In addition to that, you begin to see the government taking stakes in corporations, which back in the two thousands, people thought that China would eventually become more like the United States.
Now looks like the United States is becoming more like China, at least in that regard. And then with the tariffs, we’ve, you know, been through. Two decades of rising market concentration in the US domestically, but some exposure to foreign competition with the tariffs. Obviously, American firms are becoming less exposed to foreign competition as well, and then with these exemptions, which essentially create ways of handing out favors to firms that are sympathetic to the government, obviously that.
Means that you have a system more built on political connections, which in time makes it harder for new firms and startups to compete. And it’s again, more similar to the Chinese system. So yes, I do think there are several things to be worried about
Joanna: and you know, to put a finer point on it. Do you think that’s how progress ends in the us?
Carl: Well, I think that is the pattern that I outlined in the book. That is key concern. And so I don’t think that United States is in any way doomed. It can reverse trajectory as it has done many times, and it’s generally been a bad bet. Tibet against United States. United States has been the most dynamic economy for, well, a century and a half, but it’s also important to remember that, you know, nothing lasts forever.
And it’s striking that historically for the most part, technological leadership has been a relatively short-lived experience. I live in Britain, it used to have been a technology leader. It’s, you know, hard to believe today, but that was the case not that long ago. So things can change. They can change relatively quickly, but technologic leadership is also relative.
Position and so the United States might be helped by the fact that Europe and China is not doing particularly well either.
Joanna: Let’s pivot a little now and talk about AI and progress more specifically. One of the things that I found quite interesting and perhaps a little bit surprising until you dig into the rationale for it, was that you say that the more emphasis that we place on AI’s dangers over its advantages, the more likely it is that we’ll drift towards a monopolistic situation.
So can you explain what you mean by that and then. I guess it also raises the question of how then do we identify and isolate the signal from the noise when we are looking at discussions about risks and opportunities, but progress related to ai.
Carl: So just to be clear, I’m not saying there should be no rules and regulations around ai.
And I’m also not saying that there are no risks with artificial intelligence, but I think it’s important to keep in mind that when you create rules and regulations, you also create compliance costs, right? And we see this with the GDPR in Europe, for example, which is sort of perfect, well in intentioned regulation, but it’s smaller in young firms that have borne most of the financial burdens of that regulation. Larger companies have essentially managed to offset those compliance costs by capturing a larger share of the market. And so it’s that dynamic that we are seeing playing out with artificial intelligence and the regulation of it as well, that you create this compliance cost and then.
Obviously you might be able to control some of the risks a bit better if you have fewer players that you need oversight of. On the other hand though, you become more dependent on those few players and so much of the political power will actually reside inside those organizations, and so you may end up inadvertently cementing their position.
And so I think it’s important we think of creating rules and regulations that, you know, favor transparency and other things, but not, don’t necessarily create these significant barriers to entry, or at least we are mindful of those. Otherwise, you might end up with the market structures that you have in the pharmaceutical industry, whether it’s so expensive to take a drug to market that essentially if you are a small biotech company, you have to partner with a large pharmaceutical company in order to be able to do that.
Joanna: And in a similar vein, one of the other claims that you make in the book is that open source foundational AI models are really important to encouraging innovation. Can you explain that? And again, perhaps with reference to a historical example, there.
Carl: Yeah. So I think it, it goes back to the same point that, you know, the lower the barriers to entry are, the more people you have that can participate in innovation and the more technological trajectories it can be, uh, explored.
And we don’t know exactly what the future of AI is. Well, it may be large language models, it might also be small language models, it might be world models, it might be some fusion with symbolic ai, it might be something else, right? And so to figure that out, we need different players. Taking different bets and pursuing them.
And as long as those bets are built on the foundation of large language models, then you know these open weight models are going to be important to lower barriers to entry them.
Joanna: Recognizing that China is much more widely leaning into open weight models versus US companies that are largely focusing on proprietary models.
Do you think that then is leaning towards a situation where, if we’re looking at the lessons from history, recognizing that history isn’t a direct prediction of the future, but from history, that is leading us to conclude that China will likely be more successful given the openness versus the proprietary models.
Carl: So we saw some version of this with, you know, Japan back in the eighties, right? So the way that Toyota had Outcompeted, general Motors and Ford, Sony were Outcompeting, its Western pea and consumer electronics. And Japan was being beginning to capture a larger share of the semiconductor industry globally as well.
I think a key reason why. Uh, the predictions of Japanese that were taking it turned out to be wrong, is that we tended to emphasize well scale too much and dynamism too little, and in the end it was new firms that ended up building the software industry and e-commerce and vigorous American antitrust.
Policy played a role in that dynamism. Right now we’re seeing US firms pivoting more towards proprietary models. Metas Lama used to be, uh, leading open weight model. That space is now taken over by Chinese providers and I think many tend to look to China and see a country pursuing top down industrial policy to accomplish its objectives.
But I think many underestimate how competitive China’s industrial policy actually is. It’s not about primarily plowing resources into legacy companies, it’s having different firms competing with. Each other all across the country, sometimes supported by provincial governments, but those local leaders have an incentive to climb the party hierarchy.
And they often, you know, get promoted based on meeting growth targets. And so in Chinese politics, people behave more like firms rather than politicians in some way. And so I think there is a real risk that China has better understood this than the United States today. And that is not to suggest that China doesn’t have issue, it clearly it does. Dynamism is down in China, as well productivity growth is stagnated since the great recession. A third of the country’s gross domestic product is real estate, et cetera. But I think in this particular domain, China is currently doing better than the United States, and I think Western leaders should certainly be concerned about that.
Joanna: You don’t write about Australia in your book. It is looking at the power and interaction of nations, largely with a historical focus, but with the lessons looking forward. But listening to you speak there about how we underestimated looking at Japan and Korea, the impact and the power of dynamism, it does make me wonder whether you.
See any lessons for middle power countries like Australia in responding to the current challenges that we have around technological progress. If you were advising Australian government officials to say, this is where it would benefit you in terms of progress to focus national attention, where would you be directing people’s focus and efforts?
Carl: So I think what China’s done with open weight models, I don’t see why Australia or Europe could not do that. Right. I think, you know, emphasizing that is important because it is possible to catch up fairly quickly through that path. China has shown how that can be done. And then in addition to that, as I mentioned earlier, we don’t know exactly what the path forward in AI is, and so doubling down on the research, which is less capital intensive than scaling up large language models is something that Australia and Europe can do well.
But it obviously, it depends on attracting the. Whilst leading talent and that talent is scarce and the effect that the United States is benefiting from at the moment is hard to replicate, but there’s also a golden opportunity right now to to attract. Talent, right? If I were Australia and Europe, you know, I would be opening my arms to tropic and giving what’s reasonably happened to say, Hey, come here.
We can work with you. I think that goes for a lot of academics as well, that are, you know, tired of the situation in the United States would happily come to Europe, for example, but European institutions come match salary and that’s a real challenge as well research such institutions here don’t have the same amount of autonomy or the same amount of financial BuyPower.
Joanna: And so Carl, when you are looking at progress and positioning nations to ensure that progress doesn’t end looking in the context of ai, can you just double down a little bit on this difference between catch up versus Frontier? Innovations and how you’re seeing that play out in the context of the conversations around ai.
Because I think listening to what you’re talking about there, a lot of the opportunities you’re talking about are actually in the space of catch up. But you were briefly mentioning the deep research. I wanna crystallize this point, which is actually central to your book, that it is different structures that you need for those different points in technological progress.
Carl: So post in part in Europe. Is that, not that that Europe is less dynamic, but that has even failed to close the gap in digital the way quite different from the post-war period where Europe essentially closed gap in manufacturing and mass production. So why hasn’t Europe managed to catch up? And some people say, we need more vigorous industrial policy.
Again, look at what happened with Airbus. Essentially, you know, Europe created a peer competitor. To Boeing through vigorous industrial policy. Skeptics will say, but look at all these national efforts to see the national semiconductor industry that failed, and both are obviously right. And so what’s the difference?
I think the key difference is that when it comes to aircraft technology, and I was already. A mature industry. Back in the 1950s, the jet engine had already been developed and so Europe was catching up to relatively static target, which was Boeing, and semiconductors is much more. Dynamic AI is even more dynamic.
And as I mentioned earlier, we don’t know what the future of AI looks like. And so maybe Europe tries to catch up in large language models, but then the future of AI turns out to be world models well, but has to gain through that. And even if it turns out to be large language models, you know. These tend to develop quite rapidly, and so you’re not catching up to a static target.
And so I think in this environment, Europe would do better in focusing more on the sort of preconditions for, uh, scaling or the lack of the ability to scale, I should say, rather than trying to build new companies from scratch and in manufacturing goods. We have a single market and digital services, and that caps the return on investment.
What China and the US have in common is that they have large domestic harmonized markets that firms can scale into. And so I think priority number one for Europe is going to create a single market for services as well. And I think. Priority number one for Australia as well should be pursuing similar strategies with like-minded countries.
Joanna: Hmm. Yeah. So building out those coalitions of countries that you can have economies of scale effectively with, which is something that we’ve spoken a lot about in the podcast. And we will also speak about in the next episode where we’re going to bring together some of the middle power countries to talk about how we might actually work together.
I wanna do a little bit of a imagination experiment with you here, Carl. We’ve spoken a little bit about us and China. Imagine you’re sitting at a dinner party, you’ve got a glass of wine in your hand, and suddenly you find that that dinner party guests have just got a lot more interesting. And President Xi is now in the room.
You have one anecdote or story to tell them. What would you draw on from your book as the lesson from history that they should be paying attention to? And Trump’s visit to China actually hasn’t been postponed. And so this dinner party is actually bringing President Trump to the same table with you. So what would your story be for Trump?
But let’s start with Xi first.
Carl: I think a key sort of striking fact. From the book, is that a key reason that the first industrial revolution happened in Britain and not in Europe and not in China, is that the craft gills which were controlling production all. Over that space vanished first in Britain, and it took another 200 years for the craft guilds to disappear in China.
And that’s also the reason that, you know, industrialization was delayed by two centuries. And so it’s perfectly possible to have a few vested interests stifling progress over centuries and. The key lesson from that is obviously you need to use the state to enforce competition. And so this was done through the Turnpikes, which reduced transportation costs and created more competition between cities.
And it was also done by Parliament more directly because they realized that to be competitive in global markets, they needed to embrace factories and machines too, because British wages were relatively high. So that might not be the most, you know. Interesting story to listen to, but I think it’s an important one, to be honest.
Joanna: I love it. I think it’s an excellent story. And would you have a different one for Trump or would you still use the same?
Carl: Well, I think at the risk of repeating myself, I think the point made earlier about meritocratic civil service and the state capacity, I think, you know, that could might be a different story there, but I already told that one.
Joanna: Yeah. Look, that’s great. Thank you very much, Carl. I’ll end perhaps Carl by giving you the opportunity. You end your book by saying that actually recognizing that progress is fragile is the first step to being able to work together to ensure that we are progressing. What would your call to action to people who are listening to this podcast be?
What do you think the most important thing for individuals to be doing in this time of great change? Great progress, but also great uncertainty.
Carl: So I think it’s important to remember that an economy that’s not growing eventually becomes a zero sum game, right? So the only way for me to improve my center of living in an economy with no growth is to extract resources from somebody else, right?
It’s, uh, from somebody else’s expense. And so when you think about the kind of policies that you are supporting. Do think about them in terms of whether they’re likely to boost growth and innovation, but do also think about the redistribution consequences of them. Right? And so the British economy did grow quite significantly during the first industrial revolution, but ordinary people.
Didn’t benefit. And I think a key reason for that is also that technological change centered a lot on automation in particular, and that is not the key driver of prosperity over the long run. If all were done since 1800 was automation, we would’ve productive agriculture and cheap textiles, but we wouldn’t have antibiotics, vaccines, airplanes, cars.
Computers, right? So most growth and prosperity comes from doing new and previously inconceivable things. Those new things are most likely to do be done by startups that tend to be the one that develop new products and new kinds of industries. And so I think keeping those things in mind when you decide on the political agenda that you want to lend your support to would be.
Worthwhile doing in my view.
Joanna: And I think, um, Australia’s Minister for Industry and Science would welcome that comment. And it does resonate very much in terms of the three pillars that we have under Australia’s national AI strategy. One of which is to ensure that the benefits of artificial intelligence are widely distributed across the Australian population.
And I think. Again, this is what I mean by it’s heartening to see at least the narrative around this is really well focused and now we need to make sure that the actual policy implementation follows, which is exactly what we’re here to do at the Tech Policy Design Institute. So Carl, thank you so much for this conversation.
It’s been really the highlight of my day today, so I appreciate the conversation. Appreciate you making the time, and hopefully we’ll have you back on at some point in the future.
Carl: Thank you. It’s been a really pleasure.