Margaret Peters 0:00

Jeff, hello and welcome everybody to today's webinar. We're very excited to have Jeffrey Ding from George Washington University to present on his book technology and the rise of the great powers, which was published recently with Princeton University Press. Jeff is an assistant professor at GW. He got his PhD at Oxford University, and he's really interested in great power competition. How is it affected by new technology? There's a lot of work on AI and AI governance, and has done all sorts of interesting projects. So I'm going to turn it over to Jeff.

Jeffrey Ding 5:18

Thanks so much, Maggie for inviting me here. Really excited to present this book to you all today. Let's get into the slideshow. So I want to start with this image for two reasons. One is, I think it captures this idea of the US and China in geopolitical competition over emerging technologies like AI. But I think it also works on another level, because this type of image would never really show up on an actual chessboard. The Kings should never touch each other, and I think that illustrates the confusion we have about this topic, and we need a better understanding of how technologies affect the rise and fall of great powers. Maybe one starting point is this speech by Chinese leader Xi Jinping back in July 2018 in front of the BRICS Summit. So BRICS are Brazil, Russia, India, China and South Africa represent a large and growing portion of the world's GDP and population. And so when she goes up and talks to these emerging powers about how cutting edge technologies like AI are going to shape the development trajectory of human history. He's calling back to these past three industrial revolutions, mechanization in the first, electrification in the second, and for in the information revolution in the third. In the Chinese system, what happens is, after leaders give speeches, then you have analysts and commentators trying to decipher and interpret those remarks. So a few months later, the Chinese Communist Party publication study times goes back through that speech and analyzes it in more detail. What I want to emphasize here is they call back to those same industrial revolutions and connect it specifically to which country was able to achieve productivity leadership from those technological revolutions and taking the mantle of advanced productivity as the basis for global hegemony. So we see that same story. Those of you familiar with the International Relations literature should should see similar themes echoed there. Here's historian Paul Kennedy in this in his seminal text, the rise and fall of the great powers, where he outlines this pattern by which great powers rise and fall. First you have the changes, the technological changes, which create differentials and growth rates and lead to shifts in the global economic balances of power. And then that gradually will shape the geopolitical and military balance of power. And so for today's talk, and for the purposes of the book, I'm focused on that first step in the causal chain. How do these technological changes actually result in a shift in the global economic balance of power, specifically one great power's ability to sustain economic growth at higher rates than its rivals in the long run. The received wisdom, which I labeled the leading sector theory, is very much focused on innovation, that initial debut, the initial generation of a new technological advance. And so the story that leading sector scholars tell us that with in these times of technological revolution, one country dominates innovation in these fast growing industries that sprout off on the back of these new innovations, and they translate those monopoly profits into eventually becoming the world's most productive economy. Dan Drezner sums it up pretty well, great power requires hegemon Sebs through a new monopoly on innovation and leading sectors. And it's a pretty influential perspective here, where I work and live in DC, in terms of policy making around technological competition. Just to fill it out a little bit more, this idea is very much based off of development economist Walt Rostow work, where he identifies this classic sequence of great leading sectors, and so cotton textiles, off the back of innovations such as Hargreaves, spinning jenny, automobile sector, chemical sector, many of these fast growing industries eventually became the largest. Sector of the economy for for a period of time, for some of these countries, political political scientists have largely adopted this scheme when now we think about technologies and the rise and fall of great powers. Just to briefly preview my argument, I focus more on general purpose technologies, these foundational transformations that can't be just contained to one sector, like artificial intelligence, potentially the computer, electricity, steam engine, these types of general purpose technologies will shape economic competition. Will shape which country is able to sustain economic growth at higher rates than its rivals, and that different pathway is also going to inform the types of institutional adaptations countries will have to undertake. And I highlight specifically the role of GPT skill infrastructure, these education and training systems that widen the base of engineering skills associated with the GPT. So in the next 15 minutes or so, I'm going to get deeper into the argument then I'm going to present evidence from the second industrial revolution, which is one of the historical case studies. And then finally, we're going to talk briefly about implications for us China competition in AI. So the basis for the argument for both leading sector theory and GPT diffusion theory is that new technologies present these demands, and countries have to adapt to those demands. And for me, I think leading sector account the specific technological demands are very much focused on the initial innovation, that eureka moment that gets covered in the front pages of The Wall Street Journal and MIT Tech review. It's the most dramatic moment. It's what we're naturally drawn to. And so when we think about, okay, what are the institutional adaptations countries have to undertake? So whether that be the skill formation institutions I focus on in the book, it could also be specific technology policies. It could be the system of government, the political regime. Those adaptations are always shaped the recipes that we come up with that's shaped by Can you monopolize profits in leading sectors? Right? So maybe the focus is on investments in cutting edge R and D, who has the better industrial research and development labs, or who has the strongest intellectual property protection systems to ensure there's no technology leakage. So those types of adaptations, think the leading sector theory, the theory I'm arguing against, is very much driven by this assumption that we treat a nation like a firm with a new product innovation, and so that a country comes up with this new smartphone product, and they invested all this R and D into creating this new smartphone, and they have that brief window of time before competitors emerge with their smartphone alternatives. And it's that brief window of time when you get to monopolize profits and reap the benefits of innovation. So we'll see from this literature base in international relations, the gilpins of the worlds, the Kennedys of the worlds, Thompsons of the worlds there, when they describe and analyze how technology shapes great power competition, it's sort of like all these innovations cluster in one country, right? One firm has a product innovation. Greatest marginal stimulation to growth happens very early on. Leading sector grows really quickly. There's that brief window of monopoly profits. And I think that last quote is particularly illustrative where it shows that leading sector scholars, they trace that window until the diffusion stage, that stage in which technologies get spread throughout a population of users. For them, that's where their story stops, because that's when that window of monopoly profits ends. For me, in this book's theory and my alternative theory based on general purpose technology diffusion. That's where the story begins, the diffusion stage. But let me start with why GPTs? General Purpose Technologies, according to economists, economic historians, they've been labeled engines of growth because they often proceed huge ways of productivity growth. So So scholars have distinguished them by these three characteristics, scope for continual improvement. Oftentimes it's it's such a foundational technology that it's associated with an entire research paradigm, pervasive applicability. This capacity to be able to be diffused throughout. All different sectors of the economy and dependency on complementary innovations in all these different application sectors. So take those all together and you get this regular pattern that Stanford economist Paul David describes as this extended trajectory of gradual and protracted diffusion into widespread use across all different economic sectors. And so that suggests moving away from this innovation centers, and we have it's not sure being the first to adopt, being the first to innovate, will have some advantages when it comes to adoption, but it's not determinative. And I think this is a bit counter intuitive, but when you think about the most advanced economies, they're all going to have these frontier firms and frontier universities that are pretty well established in a broad field, like AI or electricity. And it's more about imitation. It's more about this process of taking advances that may have been incubated elsewhere and diffusing them throughout your entire economy. So for me, the institutional adaptations for to meet the demands of GPT is one of the most important. Are institutions that widen the pool of engineering talent linked to a new GPT. And it's not just about making sure the skills catch up to the technology as it races ahead, but especially that second problem identified in this slide, how to coordinate between all these different application sectors that are trying to figure out what's going on in AI or electricity, because how they adapt to that GPT is going to be dependent on ensuring that there's good information flows between the GPT sector and all these numerous application sectors, and for me that engineering knowledge, engineering skills, systematize and standardize those information flows and make it easier for a GPT to spread. So it's no surprise that after a new GPT, we often get completely new engineering disciplines. See steam engine and mechanical engineering, electricity and electrical engineering. Computer science is very much an engineering oriented discipline, even though it's not in the name. So summing up on the theory front, the main argument is, there is an alternative mechanism by which technological revolutions result in economic power transitions and GPT diffusion. We expect different things to take place. In terms of the the impact time frame, the phase of technology development, that's going to be the most important, the breadth of growth, and that's also going to shape that last column, which institutional compliments are going to be the most important. And so keep this table in mind as we're going to go through one of the historical cases, tracing which path did it take? Did it take this GPT diffusion path, or leading sector product cycle path, Second Industrial Revolution. Case, I'm going to breeze a little bit through this, but happy to go more in depth if there's any clarification questions towards the end. So the background here is, you have a technological revolution. All these new innovations are sprouting up, and you also have an economic power transition, relative decline of the UK and rise of the US and Germany. One thing I want to emphasize, it's the US that becomes the preeminent economic power, especially if you take productivity into account and all these different economic efficiency indicators show that Germany actually never surpasses the UK on those metrics, whereas the US clearly does before 1914 so that's what we're trying to explain here. So I'm going to go through all of those dimensions and just show you snippets of evidence that suggests GPT diffusion theory is more likely to be at work than the leading sector model. So on all these impact timeframes, you see a very you see this protracted trajectory that GPT diffusion would predict, especially if you're talking about the US. This all these new advances in chemicals, it's very hard to make the case that they had this near instantaneous impact, part because the US is very slow compared to Germany to establish industrial research labs. So a lot of quantitative and qualitative evidence that supports this finding that electrifications impact only becomes significant after 1914 the table of patents below, I want to highlight the mechanical patents and other patents that are containing electrical and electronic related vocabulary. You really see the huge growth coming after 1920 when electricity is spreading beyond just electrical patents into all different sectors, starting to speak that. Language and diffuse those ideas. For me, the key impact in here, the key GPT in this period, is actually incubated by machine tool advances much earlier in the 1840s in the mid 19th century, that allowed the US to adopt this interchangeable parts manufacturing system, where with better milling machines and turret lathes, you could shape and form and cut metal and wood in more precise ways to create interchangeable and standardized parts. So if a bicycle breaks, you don't need to replace the whole bicycle. You can just replace a part. And so that impact time frame figure is just showing that in the period when the US became the clear economic leader, its machine intensity and as a stand in for this interchangeable parts manufacturing process that is rising alongside it, whereas these other two technologies come much later on in the period when it comes to diffusion versus innovation, the US does not have this exclusive advantage in terms of innovation. Innovation leadership is contested, as that British Institute of Electrical Engineers quote is showing in terms of inventive genius and electrical science, US is not that far ahead, but it is ahead in terms of practical application to the industrial and social requirements of the nation, and we see that reflected in machine intensity rate as a proxy for what became known as the American system of manufacturing wasn't rooted in us as exclusive access to special innovations and machine tools. All these countries started to send inspection and study teams to the US to try to figure out what's going on, and they report back and say to the along the lines of this is about the USS advantage and diffusing special apparatus and machinery across almost every department of industry. Productivity growth, technological advances are not concentrated in just one or two leading sectors. They're distributed because it's a signal of a GPT at work, of broad based productivity growth in the US during this period, and we see this reflected in the institutional adaptations. The US is not the scientific leader during this time period. The best and brightest are going to Germany, but the US has the advantage of browning and systematizing its mechanical engineering education. Britain just simply did not produce enough as according to measures of engineering density, and Germany didn't produce the type of mechanical engineering education that was connected to practical requirements and applications. Let me just show you one piece of evidence on this quickly. The last two columns, last two rows are schools, Representative us mechanical engineering education. Top three are German schools. The US schools are devoting much more time to practical exercises in the lab and shop. So this is a Bureau of Education report that translates German engineering professor comparisons of engineering education in both countries. I'll skip over the chemical example. Just want to give you a preview of in all the historical cases. I'm only talking about the second industrial revolution here, but in the first industrial revolution, we go through the same process right, trace the technology through these three dimensions and see how that affects the institutional adaptations, similar process for Japan, the third Information Revolution case. I think the key difference here is Japan does not overtake the US, even though all the aspects of the leading sector mechanism are present. And the crucial difference here is Japan does not lead the US in the adoption and diffusion of general purpose technologies, most notably the computer and computerization. Okay, let me conclude with a few notes on how we translate the implications of GPT diffusion theory to that initial chessboard slide I started us off with US China competition in artificial intelligence, which some have deemed this era's most important general purpose technology. And I just want to show you how my findings differ, because I'm following the GPT diffusion model compared to the conventional wisdom and what you've probably been hearing overall. So leading sector model, National Security Commission on AI final report. Graham Allison, Harvard professor at the Belfer Center, Eric Schmidt, former Google CEO, who is now involved in all different aspects of us, technology policy, when it comes to impact timeframe, they're very much echoing the leading sector template right China is going to surpass the US on AI leadership. Within a decade, we're going to see the impacts of these technologies, the historical lessons and GPT diffusion theory signals something else that it's going to be this protracted and extended trajectory by which a GPT eventually makes its impact, and we shouldn't expect this near instantaneous impact. This is a marathon, not a sprint. When it comes to the phase of advantage, we oftentimes compare the US and China on innovation capacity in AI, is China investing more than the US on R and D. Do they have more valuable frontier firms? For me, it's going to be more about diffusion capacity. Who can take those advances from the frontier firms and spread them throughout the entire economy, and actually, China ranks pretty middling on a lot of those indicators. If you're interested, I had a recent article that went through this more in detail, comparing diffusion capacity and innovation capacity indicators and showing that China faces what I call a diffusion deficit, and also when it comes to the breadth of growth, is your policy concentrated on just a few different sectors like China's is, or actually this research office associated With the State Council, which is China's cabinet level body, says, Hey, maybe we should shift away from this picking winners on a few different technologies and try to support more economy wide innovation efforts. Lastly, the book argues the US is better positioned than China to take advantage of the fourth industrial revolution when it comes to institutions, to train a wider body of AI engineering talent, and also the linkages between different parts of the US innovation system. How well do your universities and industry talk and spread ideas with each other? China faces a lot of challenges in this space, and I draw out some of the implications for how the US should reorient its policy from a fortress America approach to preventing any AI technology from leaking out to a more run, faster model, where the US is focused on improving and sustaining the rate at which AI becomes adopted in a wide range of productive processes at home. Thanks for the time. Let me leave you there with some of the main takeaways. And if you're interested in the book, or if you just need a little bit more fuzzy orange colors in your life, it's available anywhere you get your books, looking forward to the discussion to follow.

Margaret Peters 27:16

Great well. Thank you so much, Jeff. This is a great introduction to the book, just a reminder, because I forgot to tell you all at the beginning that if you have a question, please put it in the Q and A section that is at the bottom of your screen. And so I'll begin with a couple questions for Jeff. So first, let me just say I found this book super interesting. I really I'm sort of like an economic history nerd, so I love this kind of stuff. I learned a lot about these different sectors, and especially about diffusion and a lot of things I didn't know. So I really want to thank you for that and for bringing technology back into IR. So first, my first question was thinking a little bit of, how do you think the Industrial Revolution and these sorts of technologies are, in many ways, they are qualitatively different from earlier outbreaks, or like new technologies that came out, but for like the history nerds who might be listening, Do you think this has anything to say about earlier new technologies, or do you think this really is about sort of the Industrial Revolution? So I was thinking a little bit about like, like the Dutch taking over and leading, sort of like the leading economic hegemon a century before the British, and thinking about, like, you know, use of joint stock corporations and things like that. So I was curious, does this have to be, you know, just like straight up technology, or can it be like ways of doing things too? So that was more like four questions at once.

Jeffrey Ding 28:55

No, it's, it's a great place to start. And I think as you read the book, and as you pointed out, Maggie, a lot of this is not my own original research. It's me taking and adapting all this great insight that's come from economic historians, historians of technology, and saying, Okay, how do we use all this new stuff that's come out and adapted to think through these, these international relations theories. So I'm glad you pointed that out in terms of, can we learn from cases before the first industrial revolution? So for some of these texts that I that I cited models key and Thompson, for instance, for instance, they do look back to the Dutch case. I think Portugal as well. For me, I use the first industrial revolution as a break point for when we started. I think it just it represents such a break point in terms of now technological innovation became more systematic and it became a more consistent. Source of differentiating between different economies, whereas in earlier periods, maybe it was much more just about what was your natural resource endowment, rather than could you improve the productivity of your economy through organized efforts to innovate? So that's why we I have not looked into those cases, but I think it could be a fascinating aspect to explore further.

Margaret Peters 30:26

Awesome, great. Okay, so my next question was thinking about, why is it the case that some countries are not just better diffusion, which you explain is really thinking about sort of education, and the role that education plays, and what's who you sort of educate. But then I was thinking about those educational institutions more, and I was wondering whether this is a democracy story or is this a specific Anglo American story, and how might we think about that? So why? You know? Because you point to a couple specific things, like different sort of more open groups in the first revolution, and then thinking more about these types of universities that come through, especially the land grant universities that come through later, and thinking about, is this about democracy, or is it about culture? Or could I adopt if I wanted to be the next winner? If I'm Indian, I want to be the next winner? How should I think about these things?

Jeffrey Ding 31:34

Yeah, so the from the text in the book, the answer is, I don't go that next step in the causal chain, so I leave it as the deepest I go. In terms of the cause is, which country is able to develop this GPT skill infrastructure, these education training institutions that widen the pool of engineering talent. For me personally, I don't think there is sort of a silver bullet or sort of one size fits all prescription. So India, this is how you build better skill infrastructure, in part, because it's hard to predict what is going to be the next GPT as well. You pointed to some of the things, right? So sometimes it is a very top down industrial policy, even if by accident. I think the Moral Act that that expanded all these land grant colleges for the agricultural and mechanical arts was really crucial to the US ability to expand its pool of mechanical engineering talent. But it's not like the US actually went in with it saying this is an industrial policy to adapt to the second industrial revolution. I think generally, there's been good evidence that more decentralized systems, so not necessarily democracy versus authoritarian, but just more decentralized approaches to science and technology do correlate with higher rates of diffusion in other technologies, and I also think that that might allow for your higher education institutions to adapt more organically and naturally to changes in technology. So in the Japan case, one of the issues with Japan's ability to cultivate this computer science, this software engineering discipline, is the their ministry of education tried to coordinate this from the top down and establish just a few sort of like centers of excellence in Software Engineering Education, whereas the US had this more decentralized approach to adaption, to adapting to where computer science was going. So those are a few starting points where, if I were to kind of go one step further down to figure out what causes variation in GPT skill infrastructure across countries, those would be a few starting points.

Margaret Peters 34:02

Awesome. Yeah, that'd be, think that's great, especially for our grad students who are on the call, who might be thinking about where to take this sort of topic. Next, I'll ask you one more, and then I'll open it up to questions from the audience. So my other question was, in this book, you focus a lot on the role of education and diffusion of technology. But then, as somebody who thinks a lot about firms, I was thinking about, are there technologies or problems where, like some things, the leading firm or the one who is producing the technology wants greater diffusion. And so we see that versus others, where they try to stop it, and like, how does that affect competition and diffusion? So I was thinking, you know, some some industries like electricity, of course, you want more customers. Do you want everybody to adopt it. But then I could imagine, even you know, thinking about in some ways, like you do want everybody to adopt AI, but like you want them to adopt my AI and not necessarily create their own AI or not improve upon it. So how have you thought about the industrial incentives here?

Speaker 1 35:28

Yeah, so a few things. Just you mentioned your thinking on firms, and I was just trying to bounce off of that in terms of what is the role of firms and in terms of providing this GPT skill infrastructure, right? So the maybe there's some logic to this type of talent pool is this like public good? So individual firms were under provided, and you've shown today that firms are not lobbying as much for importing foreign talent and sort of benefiting from those pathways I show in the Japan case, that the US, that's one reason the US was better equipped than Japan was us. Was able to benefit from those pathways. In terms of your specific question on how, kind of the incentives for firms on more widespread adoption? I think it's a tricky question. It's another area that I wish I could, could have explored more. I think the closest I get to was thinking about Microsoft and Intel in the in the computerization case, and there a lot of people were coming up with this portmanteau, this wintalism, right? So the US leadership in this space comes because we have just the two dominant architectures in the Intel central processing unit and the Windows operating system, and that's what shapes the computerization trajectory. And I guess there you would think that just more people using computers is good for all of the firms. Um, one thing that I found was it's not necessarily the case that just having the dominant players get their standards set up and kind of locking everyone out, that might not lead to the most sustainable diffusion of of computerization, and actually, like some of the US, government's efforts, anti trust efforts, were really critical to making sure other players could compete, and you weren't just locked into the wintalism system. So I do think there's analogies here in AI as well as you were saying currently. It seems like open AI is the dominant player. It's like use our system. I think there's a world where my book suggests for the US to really unlock this GPT diffusion effect. There needs to be a world where open source AI models can thrive and compete alongside the open AIS of the world and the anthropics of the world. So, so that's, that's my initial hypothesis on that line,

Margaret Peters 38:29

Great. I'm going to take my perogative and ask you another question, because I can. So this got me thinking about so the Burkle Center had an interesting talk recently with the Commander of the Space Force, which was super interesting. So then this got me thinking about some of these technologies where, like, the startup costs are super high, but are really important. So like, I know you sort of focus a lot on AI, but then I was thinking, because I'd gone to this thing about Space Force recently, which was super interesting, about the role of space and communications, and is there a problem with some of these other technologies where you have, like, really, really high startup costs that just don't allow them to diffuse? And could that so this is sort of like pivoting back to leading sector theory, like, if, for some reason we, like, found that, like, there was something awesome in space now that there isn't anything awesome in space, but like, particularly awesome in space is there is, Are there cases like that where, like, the startup cost of diffusion are just so high that it does allow one leading great power to sort of key in on the market.

Jeffrey Ding 39:49

Yeah. So this was one of the strands I was trying to get at a little bit in the book, and one of the alternative explanations. Was, do you need the military to be that initial sponsor of these gpts? And I think that is the story we often tell, is sort of the military provides this, this stable source of demand, of procurement, and then handles these large startup costs you're talking about, and then the technology can take off from there. I think there's some credence to that argument. A few things I'll point out that's not that's definitely not the story for electricity. So it's not always the case. It's not really the case for AI, I think there was some, there was a certain amount of investment from DARPA into like the three big US AI universities, but that was much earlier, and it wasn't really connected to this new deep learning paradigm. And I think the what I'll say on this front is actually one more example, even with the machine tools and interchangeable parts manufacturer. A lot of that was incubated at the US, US armories. So there is a role for like, maybe government and military to play, I will say, though, that's very much focused on the innovation stage right, who can start the trajectory, and then actually, when the technology is diffusing, when a technology has a potential, diffuse across all different civilian sectors, then you'll see the requirements diverge pretty sharply from what the military wants out of the technology, or what the government wants out of the technology. And so sometimes too much involvement from the military the government, at that beginning will make will actually prevent you from shifting towards the commercial trajectory. So I think it's an important factor, but I don't necessarily see it as particularly significant.

Margaret Peters 41:49

Great. All right, let me start with some of the questions from the audience. So one question is, is there a historical pattern in which diffusion matters less over time, because the emergence of new technologies is increasingly endogenous to already technologically advanced countries and those societies that already have the skill infrastructure. So we could think about that a little bit with like AI in that in the last computer revolution, the US developed all of these, you know, university institutions that you know, even you know our students, you know that we see at our universities are probably much more technologically savvy than students at other universities, just because they've, like, grown up with all this same technology. I do feel bad because we told them all, like, a decade ago to learn to code, and now, like that doesn't matter. But, you know, thinking about about that so is there's a case, especially in the this new industrial revolution, where can you so both? Is it endogenous? But then I'm going to add on a little bit like, Can other countries catch up?

Jeffrey Ding 43:04

Yeah, yeah. So Arthur, it's a great question in terms of, maybe now the already technologically advanced countries are set up to continue to produce and win in these new technologies. And the way Maggie phrased it in terms of, is there any ability for these rising powers to catch up? So I would say, for me, the one thing is, I'm looking at the set of kind of already pretty advanced economies. So I'm trying to look at, okay, yes, these advanced economies are the ones that are producing these new technologies. But then who actually wins out from these new technologies? So in all the historical case studies, we are only studying the countries that have all these firms and universities at the technological frontier. The other point I'll I'll emphasize here, is one of the most interesting things I came across in my research was there was this, I think it was an OECD study. Andrews et al is the source, if people want to dig into it. But they found that the initial, the initial adoption gap is shrinking. And so by that, what they meant is the time that when a frontier firm in one of these technologically advanced countries you mentioned in your question, by the time one of those firms comes out with a new innovation and a firm in another advanced country adopts it, that gap is shrinking due to globalization, all these different things, that all these different factors, and also one of those factors is like multinational corporations setting up research labs in other advanced countries, and so that gap is shrinking. But what that study found was. So another gap was rising, which was the intensive adoption gap, which is once a frontier firm in your country, introduces, or first adopts an innovation, and the time it takes for a certain percentage of your small and medium businesses to adopt that innovation and for it to spread throughout the country. That gap is actually increasing over time. So for me, that suggests, okay, now all these different advanced, technologically advanced countries are going to have the opportunity to absorb new innovations quickly to their to their top firms. And now what really differentiates countries is going to be the race from to get those innovations from the top firms throughout the country. So it actually, for me, that lends more credence to there is an opportunity for countries to catch up, even if they're not the initial innovator in this space,

Margaret Peters 46:03

Awesome, just as a follow on then. So leading sector theory would sort of say, like, the government should invest a lot in these, like, first new, newcomer technologies, or what they think is going to be the next thing would your suggestion be, instead, like, let's say, you know, Modi picks up the phone and is like, what should I do? Or even, you know, I don't know, Brazil, the Brazilians pick up the phone Lulu. Lulu picks up the phone and calls you. So I was curious, you know, would you just suggest to them, like, what you should be doing is focusing on, like, if you had a budget to spend, would you spend it? Then on helping firms adopt these new technologies and or education, or both, or like, where would you put your money?

Speaker 1 46:54

Yeah, so the first the caveat to this is, this is not meant, sort of. The disclaimer is, this is not meant for, like all purpose solution to technology strategy, right? This is if you are optimizing for how to adapt to emerging technologies and sustain growth in the long term, here's what you optimize for all these countries. You mentioned, India, Brazil, they're gonna have to adapt to other interpretations of technological leadership, and that could be like making sure you have supply chain and independence in a few strategic sectors. So there, the leading sector model might be more appropriate other you know, if you're maybe only focusing on military power there, you might pursue a different set of strategies, but, but sort of, if you're in it for what I think is the most important aspect of AI and all these emerging technologies, which is, in the long run, productivity growth is all that matters and and a lot of people have shown that it matters significantly for power of all forms. Then, yes, I do think it's about this broad based, diffusion centric strategy. Maybe one thing To illustrate this, more concretely, is with the recent Foreign Affairs piece, I looked at the chips and science act, and if you separate out the two main planks of the chips and science act, it maps pretty neatly onto the leading sector model and the GPT diffusion model. So all the stuff that gets the news and the headlines is the boost domestic chip production we got to give the Intels of the world make sure they can compete in this industry, this really fast, fast moving, competitive industry. Then there was all this stuff about STEM workforce development and a more broad based approach to upskilling people, to diffuse new technologies and give you one guess as to which plank has the money that's been authorized has been allocated and implemented more effectively, right? So I think for these countries, for India and Brazil, be wary of these technology policies that favor like a few concentrated interests. Diffusion policy is really hard to do because the benefits, by definition, are so dispersed and distributed and so yeah, that that would be my advice for these potential rising powers.

Margaret Peters 49:29

Great. That segues next into another good audience question. So thinking about the chips and sciences act and you know, things like that. Who wins the presidency next month? Do you think that? How do you think that might affect, you know, it's everybody on all of our minds all the time. How do you think that might affect this latest new industrial revolution with AI, if at all?

Jeffrey Ding 49:58

Yeah. So, for me when it comes to technology policy, and especially technological competition with China. What's interesting to me is we almost do have a bipartisan consensus on this, and it's very much influenced by the leading sector model. If you look at some of the most prominent policies, right the october 2022 Biden administration, export controls, on high end, AI chips, that is a very drastic policy, almost, I think, as important as the chips and science act in terms of this is the closest we've gotten to an economic containment policy, if you ask people, the state of justification of that export control policy on a high chips, is to slow military modernization, but then they'll very quickly pivot towards this is to ensure we have leadership on our frontier technology. So I think actually across potential administrations, they would have that same approach towards we need to our resources, our political capital is focused on sort of a fortress America like approach towards competing with China. And for me, I think that there's just a much better way to invest our resources in terms of a more effective approach to competition. And I think just for me, I have a different definition of what technological leadership in AI actually means. So this the short answer to the question is, maybe sort of the particular political party, the particular candidate, might not shift us Technology Policy all that much.

Yeah, although they might shift funding for universities,

yes, yes

Margaret Peters 51:49

Not to be self interested here.

Jeffrey Ding 51:51

Yeah, I've been trying to actually bracket, stay away from following that too closely. But yes, good point.

Margaret Peters 52:00

All right, in the interest of time, I'm sort of going to combine these next two questions, because I think they actually work together. So one question was about, you know, the costs and benefits of these technologies and the winners and losers. So we know, with new technologies, you've always had some people left behind. The next question focuses on in spatial economic terms, dispersion versus agglomeration. And the reason I'm putting these together is because if we look at the last the like sort of computerization last 30 years, what we've seen is the winners are typically spatially concentrated in our large cities throughout the West. You know, if we think about US or Europe, it's the large cities or the winners, and then the more rural areas that left behind. And there are winners and losers inside cities too. But like generally, if you've been in a city, you've done well if you've been in more rural area. So do you think that? So this thinks a little bit both about the spatial economic terms of like, do these inherently move to sort of agglomeration or and then also thinking about dealing with the winners and losers.

Jeffrey Ding 53:17

Yeah, this is great. So really, really important questions. I think it touches on. We could go in so many different directions on this, um, you know, I think there is a world where you can get the maybe there's a sort of one way to phrase this question is, can you get the impacts of GPT diffusion? Can you get that economy wide productivity boost from just spreading electricity and computers to all the big cities and sort of these clusters? Um, I would want to study this more in detail, because I know the US actually invested a lot in, like, rural electrification programs, and there's a lot of interesting dynamics in China right now, where, whenever I present this to a more China studies audience, they're like, if we can just get AI diffused across all the coastal cities and the first year, cities isn't that enough. I think I would lean more towards it's not enough. You do need that, those dispersion forces. You do need more broad based productivity growth. And I think that does come from not just this diffusing to the the sort of the clusters, the innovation clusters. And I think in the book I write, it's about, can you get the stuff from Silicon Valley to, like, Iowa City, which is where I grew up, and so I don't have the, I wish I want to go back and get the stats on, like, Okay. How broad based was electricity diffusion by this time period, or how broad based was computerization and how kind of how much dispersion Did you really need to get that sustained productivity growth? But yeah, in the China context, for me, I think that's one of the challenges they face, is they're certainly benefiting from all the agglomeration effects and developing these clusters, right? Zhong Guan song is not John Guan, and Beijing is not that far behind Silicon Valley, but they haven't been able to bridge it from those centers and spread it across into like the inland provinces or the fir or not even to mention like the further western provinces in China.

Margaret Peters 55:44

Yeah. And then one last thought I had, I guess he's not your colleague any longer. I say a little bit about, like, weaponized interdependence, and that argument

Jeffrey Ding 55:58

We still claim him

Margaret Peters 56:02

I was thinking, I was like, wait a sec, yes. So then I was thinking a little bit about, I guess it also ties into, like, if you are talking to the powers that be in Washington or elsewhere, is using the like, network power. So because the top AI model is in the United States, would you say you should limit this to just the United States, or would you argue that, you know, actually it's all just about diffusion. And what we want is we actually want foreigners working on these models and making them better, but what we want to be able to do is have our scientists and everybody every day learn how to use these tools to the best. Yeah,

Jeffrey Ding 56:50

I think this is a great thread that sort of differentiates a GPT from a technology that's a GPT from a domain that's more suited for weaponized interdependence, or for this leading sector approach. So I think the chip controls the the reason why weaponized interdependence can can justify chip controls is it's an it's an industry where the there's sort of like a concentrated chokehold, and there's just really one or two firms sometimes in some of these key semiconductor components, whether it's the semiconductor manufacturing equipment or whether it's the firms that are designing some of the highest end chips, like Nvidia. And so there it's, it's sort of, there is a rationale for using weaponized interdependence for AI, yes, the US has the best models, but a lot of the models are also open source, and Chinese models are not that far behind. And so it's sort of like, it's like you wouldn't really expect for something as broad of a domain as electricity or computers, for there only to be sort of one competitor or one locus for, for, for these concentrated chokehold effects. So that's why I don't really see that as an effective approach when it comes to AI, especially if you take long run considerations into account, right? Any choke hold might hold for like, the next four or five years. But if we're talking in terms of when gpts make their mark, by that time, you're gonna have a lot of different model providers, and so I think the salience of of those weaponized interdependence mechanisms will be reduced.

Margaret Peters 58:52

Awesome, great. Well, I want to thank you so much for coming and presenting this book. It's a really for all of those who are on the line, it's a really interesting read. I highly recommend it. You see the link in the Princeton, the Princeton University Press link, you know, throw Jeff that $1 he will get 50 cents or whatever. You know, the huge amount of money you're gonna make up. It's a super interesting read, and it really got me thinking about the role of technology. And I'll say, like, in many ways, it was kind of hopeful, in the sense of, I think there's so much fear right now about AI, and, like, the our competition with China, and I feel like, you know, after I read it, I felt like we're gonna be fine. Like, you know, we're gonna be okay, like, we have all these great systems in place. So I think it is very hopeful message for all of us who care about these sorts of issues. And I just want to highlight for those of. You who might be interested. Our next talk like this is going to be on November 13. Joshua Busby thinking about states and nature, the effect of climate change on security. So similar sort of topic, we're going to move to thinking a little about climate change and security instead of technology and security. But again, thank you so much for coming on and talking about your fabulous book.

Jeffrey Ding 1:00:22

Thanks so much. Thanks for having me.

Margaret Peters 1:00:24

All right. Thank you.

Transcribed by https://otter.ai