---

Jeffrey Ding is an assistant professor of political science at George Washington University. He is the author of a new book on emerging technologies and great power competition, called Technology and the Rise of Great Powers, as well as author of the ChinAI newsletter, an indispensable source for news and analysis on developments in China’s AI industry. In this lightly edited Q&A, we discuss the provocative argument in Ding’s book that political leaders are fixating on the wrong things in the race to get ahead in AI, and that the U.S., not China, is better at diffusing key technologies.

Q: The main focus of your book is “general-purpose technologies” (GPTs) and how their spread determines the rise and fall of great powers. Could you explain what a GPT is? How do we know when something is a general purpose technology, and why is AI one? 

A: General purpose technologies (GPTs) are fundamental advances that have the potential to transform countless sectors of the economy. They have three characteristics: first, they have a scope for continual improvement. Second, they are pervasive, in the sense that there’s a wide variety of use cases. And third, they are characterized by technological complementarities, which means the full impact of a general purpose technology is only unleashed when other complementary technologies in various sectors also transform and accommodate the new GPT. Taken together, they’re often deemed to be engines of growth, because the arrival of a GPT like electricity, the steam engine, or the computer — and maybe AI — often precedes a wave of productivity growth.

It’s a difficult exercise to pinpoint a GPT before it arrives and before it makes its mark. Because general purpose technologies have a huge scope for continual improvement, you sometimes don’t realize which applications are going to make the most impact in terms of productivity benefits. Think back to the advent of electricity: Some of the early applications were in lighting, and that was obviously a game changer for a lot of reasons. But the most important benefits to productivity came from much later applications in factory electrification, and also the ability to just use a bunch of different appliances that were connected to electricity. 

There are some early indicators that suggest out of all the emerging technologies, AI has the greatest potential to be a GPT. One way of tracking the “GPT-ness” of certain technology fields is by looking at patent citation patterns. When patents of a particular technology like AI, versus biotechnology versus blockchain technology, are cited by a broader range of other patents in different technology classes, it gives a signal of how broad the applicability of a technological field is. People have done the same exercise using job posting data to show the breadth of different industries that are looking for people with knowledge of AI. Across those metrics there’s strong evidence that AI has greater potential to be a GPT than other enabling technologies such as biotechnology and blockchain.

A good way to think about it is, if you’re trying to think about the difference between automotive technology and artificial intelligence for instance: are innovations in AI changing the field of automotives more than the field of automotives is changing the field of AI? In that case, automotives are closely following and have to adapt to what’s happening in AI. The reverse is not as true, meaning AI is more likely to be the general purpose technology. You could make that same comparison for other candidate technologies like robotics and biotechnology.

You argue in your book that political leaders tend to focus on the wrong thing when it comes to technological change, fixating on breakthroughs and innovation capacity, rather than diffusion. Could you explain what you mean by that?

It’s helpful to break down technological change into three phases. First is invention, which is the very first breakthrough. Second is innovation, which is the first practical application of that invention. And then the third is diffusion, which is the adoption and spread of that innovation throughout a population of users. Oftentimes in our analysis of how new technologies shape international politics, we stop at the innovation phase. It’s about who is the first to invent and then debut a new application or process. 

My book is centered on what comes next, and how that new product or process gets adopted into productive processes —how the new technology that gets incubated in a frontier lab or university filters throughout the entire economy to get to a small firm in Iowa, for instance, where I’m from.

For some technologies, that innovation phase might be really important, like new pharmaceutical treatments. It’s very much about which firm can come up with that newest innovation. But for general purpose technologies, the most important factor is always going to be diffusion capacity and a country’s ability to adopt and spread these GPTs across their entire national economy.

Your book looks at three case studies of how a general purpose technology catalyzed a country’s ascent. Maybe we could start by talking about those three instances, and what was the diffusion advantage in each? 

These three case studies map onto what some scholars deem the three technological revolutions. The first industrial revolution was between 1760 and 1840, when Britain solidified its economic preeminence and became the global hegemon. The key general purpose technology in my analysis, was the spread of access to cheap iron, which enabled early stage mechanization. 

…at the highest levels, up to Xi Jinping, Chinese leadership has pointed to these past industrial revolutions, saying that now China needs to compete in this fourth industrial revolution driven by AI. 

The second industrial revolution was when the U.S. took the mantle of economic leadership from Britain between 1870-1914 . My argument here is that the U.S. was very far from the scientific research frontier. Other countries, such as Britain and Germany, had the advantage when it came to new innovations in the chemical sector. The U.S.’s main advantage was in adopting interchangeable parts manufacturing, which was a product of new advances in machine tools that made it so that you could more precisely shape and cut wood and metal parts. So if a musket or a rifle breaks down, for example, you don’t have to buy a whole new rifle. You can just buy the part. 

This came to be known as the American system of manufacturing. These advances in interchangeable parts affected industries from sewing machines to bicycles. The U.S.’s advantage was not necessarily in making better or more advanced machines. It was the eagerness and the intensity with which the U.S. economy diffused this American system of manufacturing.

In the third industrial revolution, between 1960 and the start of the 21st century, Japan had an opportunity to leapfrog the established technological leader and take that mantle of economic leadership, based off of its dominance of new industries, such as consumer electronics, HD TVs and key semiconductor components. But it didn’t come to pass. What I found is that even though Japan seized a commanding market share in all these sectors, it was not as successful as the U.S. in adopting and diffusing the key general purpose technology of that time, the computer. Japan fell far behind in terms of computerization rates and ultimately never sustained productivity growth at high enough rates to overtake the U.S..

Is there something to be said about the export orientedness of Japan’s economy here, in the sense of its focus on exporting electronics to the world, but not necessarily adopting them at home?

There’s a lot of validity to that point. The main theory that I’m arguing against is referred to in international relations as leading sector theory. That approach almost treats national economies as firms with product innovations. So you’re a country-firm, and let’s say you come up with a new iPhone: you have a brief window of time to export and sell it all around the world and make monopoly profits. But over time, other countries come up with their own version of the smartphone and that window fades. For leading sector theorists, it’s all about that brief window where you can monopolize exports and make all these excess profits from the export so you can reinvest that into the economy, leading to sustained productivity growth. 

People saw Japanese firms dominating these export markets in consumer electronics in a very visible way, triggering fear of the Japanese economy overtaking ours. The behind the scenes, gradual process of a small Japanese firm adopting computerized information management systems for the first time — that’s never going to make as much news. But I would argue it is just as meaningful as the fact that Japanese consumer electronics were very popular around the world, especially when it comes to productivity growth, which historically has been a key factor in the rise and fall of great powers. 

How do Chinese leaders think about innovation? Does China’s AI strategy adhere to this traditional leading sector model? 

What’s really interesting about Chinese writing and thinking on this subject is that at the highest levels, up to Xi Jinping, Chinese leadership has pointed to these past industrial revolutions, saying that now China needs to compete in this fourth industrial revolution driven by AI. 

But I think Chinese leadership has learned the wrong lessons from these past industrial revolutions. Their technology policy is very much rooted in that leading sector template, which is preoccupied with innovation capacity and making sure that China is a world leading innovation center in AI. When you look at the Chinese government’s benchmarks and targets, they have been much more successful at hitting goals when it comes to R&D spending as compared to education spending. The former is much more optimized around coming up with new innovations. The latter is much more about creating a broad base of talent that’s necessary for adopting and integrating general purpose technologies like AI. 

In the book I write about the example of Yuan Wenkai, who’s in charge of 4PX Express Logistics warehouse, which is a logistics firm in which Alibaba has a controlling stake. He’s someone who graduated from an ordinary Guangdong vocational school and has now become an expert in automation management. When Alibaba went public on the Hong Kong Stock Exchange, he was one of the people there ringing the gong. Yuan Wenkai is not someone who’s going to be coming up with new paradigms in AI like Andrew Ng or Qi Lu, who are two big names who generated a lot of tension about Chinese firms poaching top tier talents from Western firms. But average technical staff like Yuan, who can manage automation processes are the type of talent that I’m most interested in when it comes to measuring and assessing a country’s diffusion capacity in AI. 

It speaks to the crux of the argument, which is that when it comes to which country is better prepared for AI in a diffusion centered framework, it’s about the ability to train a wide pool of average AI engineers, not necessarily about which country can cultivate the best and the brightest.

Many people assume that China is adept at commercializing technology, while the U.S. is great at inventions and breakthroughs, but slow on bringing them to market. Look, for example at commercial drones, or LFP batteries for electric vehicles. But you argue the opposite. Why is that? 

Left: A NIO Power electric vehicle charging and battery replacement station in Shenzhen. Right: The Fuxing high-speed railway in Tianjin Station. Credit: Markus Mainka, Andrey via Adobe Stock

The conventional assumption certainly is that China has a stronger capacity for diffusion than the U.S.. But a lot of those evaluations are based on a few striking examples. You mentioned a few, but another that comes to mind is high speed rail and the widespread scaling up of infrastructure. I do think that China has advantages when it comes to adopting certain kinds of technologies, especially ones that the government can push for from top down, as well as consumer facing technology, where the large population helps to scale usage and bring costs down, such as with mobile payments or food delivery apps. 

But when it comes to general purpose technologies, productivity growth comes from businesses adopting process innovations like AI, cloud computing, computers, and even just the internet. There are a few examples that counter the assumption that China is ahead. One indicator is that China lags far behind the U.S. when it comes to adoption rates of computers, industrial software and cloud computing, all of which are key enabling technologies for artificial intelligence. 

A more exhaustive exercise I did was to look at different science and technology indicators that global indexes collect and sorted them into the ones that align with innovation capacity — such as the research reputation of a country’s top three universities, or the average R&D spending of a country’s top three firms — and ones that align with diffusion capacity. For example, indicators of the strength and robustness of university to industry linkages, which are important for diffusion capacity because you want different sectors of the science and technology system to be talking to each other and spreading ideas. China’s linkages are much weaker than those of the U.S.. Overall, on innovation capacity indicators, China ranked around top 10 in the world, whereas on diffusion capacity indicators, China’s average ranking was closer to the 40s. So those are a few pieces of evidence against the conventional wisdom. 

Lastly, specifically about the talent base in AI, I used one metric on the number of universities that have at least one researcher who has published research at an established AI conference. It’s a very low quality baseline, these aren’t the best of the best universities. China only has 29 of those universities, whereas the US has 159. So once you get past the Tsinghuas and Peking Universities of the world, which are leading centers of innovation in AI, the bench is not that deep, whereas the U.S. has a much broader pool of institutions that can cultivate talent necessary to diffuse AI at scale.

What are the specific impediments to China’s diffusion capacity compared to a place like the U.S.? You’ve argued that planned economies limit the ability for these new advances to permeate and spread. Is that true for China? 

One of the simpler explanations is just that wealthier countries adopt new innovations faster than less wealthy countries. China sometimes does get ahead in diffusion when you can mandate diffusion because you’re willing to eat the costs, like high speed rail — public goods that the Chinese government provides, even if it’s not necessarily economically efficient by market processes. As for consumer-facing apps that are very low cost, especially when legacy infrastructure like credit cards does not exist, those can also spread very quickly. 

…there are elements of recent Chinese policies that signal the central government taking a more active role in managing technology companies and technology policy. That’s where it could actually hamper the adoption of AI at scale…

But when it comes to something like the decision to go on the cloud or not, or purchase a subscription for a very expensive piece of industrial software, that has to be a market-based process, that has to be profitable to the Chinese firm. In wealthier countries, the firms just have more money to spend. In less developed countries, if we’re talking about firms in inland provinces, they have less of a budget to spend on some of these things. Now with each specific technology, like cloud computing, we can get into the details about things like data privacy and monopoly concerns, but overall I think it’s an important structural point that GDP per capita is often correlated with higher adoption of new technologies.

But to go back to your third industrial revolution example, a country like Japan had relatively high GDP per capita though, did it not? 

There are many different factors that will affect a country’s ability to adopt and diffuse general purpose technologies. One is just the average level of economic development. Other factors include the extent to which a national market follows the same technology standards, which makes it easier for GPTs to diffuse. Another factor is the level of competition. If there isn’t effective competition, and a few really strong firms dominate the market, that reduces the incentives and the pace of general-purpose technology diffusion. 

For the purposes of my book, I highlighted skill formation and the ability to train the talent needed to catch up and keep pace with the technology, in part because it permeates everything. Talent is the primary resource when it comes to modern technology development. So going back to the Japan case, maybe it was closer to the U.S. on some of these other factors, but Japan was not able to train and attract a wide base of software engineering talent and was significantly outpaced by the U.S. on that front. 

Can a country like China, with a strong, centralized government, accelerate technology diffusion through policy alone?

China is in some ways moving in the right direction when it comes to science and technology policy, not necessarily because it has all these flashy policies to develop AI or biotech from the top down, but more so because China has slowly and gradually moved towards more decentralization of science and technology policy, where a lot of decisions are being left to provincial and local governments. 

We’ve seen from historical cases that a decentralized approach is really important when it comes to general purpose technologies, because the paradigm is constantly shifting. If you were making AI policy just three years ago, it might be structured around computer vision, whereas if you’re trying to make central AI policy today, it would be almost completely shaped by large language models. It’s very hard for a central government to deal with these continually moving GPTs. So China’s overall shift towards a more decentralized approach is a good thing. 

At the same time, there are elements of recent Chinese policies that signal the central government taking a more active role in managing technology companies and technology policy. That’s where it could actually hamper the adoption of AI at scale, because then you have the central government trying to pick winners and choose certain technology trajectories that might not be the most fruitful. Usually it’s the actors on the ground closer to what’s happening that have a better sense where technology is moving.

Did writing the ChinAI newsletter help to inform your book’s research?

I started the newsletter back in 2017 when I was an intern at the Centre for the Governance of AI, before it was popular to talk about China’s AI development or research. I started by publishing translations of this book by the Tencent Research Institute about China’s national AI strategy after finding that nobody was analyzing it in English. 

The basic premise of the newsletter is that there’s a lot of interesting ideas and insights coming from Chinese people writing about China’s AI landscape that we don’t see from English language coverage. A lot of the ideas in my book about why China is behind on diffusion capacity actually came mostly from Chinese language analysis about where China stands. There was very little English language coverage on this, so I don’t think I would have come to those ideas if I was only reading English language sources. So in that way, doing the newsletter was essential to coming up with some of the ideas for the book.

---