Winning the Marathon, Not the Sprint: An Interview with Technology and the Rise of Great Powers Author Jeffrey Ding

In recent years, the techno-geopolitical competition between the United States and China has intensified. Both nations view winning the “Fourth Industrial Revolution” as critical to their global standing, with AI innovation emerging as the key battleground. Chinese international relations scholar Jin Canrong argues that China is better positioned to secure victory, while the US Congress’s National Security Commission on Artificial Intelligence warns that America risks losing its technological leadership without adequate preparation.

Jeffrey Ding, a rising scholar based in Washington, D.C., offers a different perspective. In his new book, Technology and the Rise of Great Powers, Ding argues that great power competition isn’t about who dominates innovation or who first develops technologies like ChatGPT. Instead, it hinges on who can effectively adopt and diffuse these AI innovations across industries. Over the long term, the widespread diffusion of such general-purpose technologies (GPTs) is what drives economic growth.

From this lens, Ding sees the US-China AI competition differently. He believes the US has a clear edge, thanks to its superior capacity to diffuse AI applications across various sectors. However, he criticizes current US policies for being overly focused on restricting technology outflows. This short-term emphasis on innovation leadership, he contends, overlooks the more important objective of fostering AI diffusion––which is a significant strategic misstep in his view.

Ding’s overarching message is clear: technological revolutions are not 100-meter sprints but marathons. Winning this long-term competition requires a steady and robust supply of AI talent. In Taiwan, initiatives such as the National Science and Technology Council’s “AI for All Industries” project and the Taiwan AI Academy echo this approach, aiming to expand the talent pool and enable various industries to integrate AI, thereby boosting economic performance.

At DSET, we closely track issues at the intersection of AI governance, techno-geopolitics, and economic competition. To dive deeper into these topics, You-Hao Lai, our Washington-based researcher, met with Professor Ding to explore his perspectives on US-China AI competition and discuss potential directions for Taiwan’s AI policy. Here’s the full interview.

Innovation or Diffusion? Unlocking the Key to Great Power Competition

YouHao Lai (L): In your book, you compare two paradigms that explain how technological change influences the rise and fall of economic powers: the “Leading Sector” (LS) theory and the “General-Purpose Technology Diffusion” (GPT diffusion) theory. You argue that the key to the ascent and decline of great powers lies not in innovation within a few leading sectors but in the ability of general-purpose technologies to diffuse across industries and permeate diverse economic activities.

Before diving deeper into your book, could you briefly introduce yourself and share what motivated you to write it? What inspired your focus on technology issues and US-China economic competition?

Jeffrey Ding (D): Yeah. So I got interested in this field as a master student at the University of Oxford, where I was an intern at the Center for the governance of AI. At that time, my main focus was studying China’s development of artificial intelligence, and so my research interests were in the US-China competition in emerging technologies like AI. But as I was pursuing my PhD dissertation at Oxford, what I found is there was a lot of speculation about who was going to become the leader in AI, but so much of that was unfounded, ungrounded speculation. And so I wanted to dig deeper into the history of past technological revolutions and how they affected the rise of great powers to understand if there were any lessons we could learn for today’s great power competition in AI.

L: So your academic journey at Oxford was the starting point. Next, I’d like to clarify some key concepts in your book. Could you briefly explain what “Leading Sector” and “General-Purpose Technology” mean? How can we distinguish between the two? When applied to AI, would you say it falls under a leading sector, a general-purpose technology, or perhaps both?

D:  The two mechanisms in the book are leading sector theory and general purpose technology diffusion theory. I define leading sectors as new fast growing industries that sprout up off the back of technological breakthroughs. So historically, some of the classic examples of leading sectors are the steel industry, the canal industry and the automobile industry. I define general purpose technologies as foundational transformations that can’t be contained to just one sector or one industry, because they have the potential to diffuse and spread across a countless range of economic sectors. So these are technologies like electricity, the steam engine and the computer. 

When it comes to AI, I think the key difference for me is not necessarily whether something is a leading sector or a GPT. It’s about how it could reshape economic competition between great powers. In the context of AI, I think the leading sector pathway says it’s all about which country can monopolize innovations in the AI industry, and that it’s about that early window of time when the new innovations appear, and who can capitalize on that brief window by which leading sectors make their mark. The GPT diffusion pathway says it’s more about which country can adopt and diffuse AI at scale throughout its entire economy. That process takes decades. It’s not about which country has a monopoly on innovation. No one country can actually control all innovations in a GPT. So the difference is they suggest different pathways or mechanisms by which countries can take advantage of emerging technologies to achieve economic leadership.

L: So whether AI is a leading sector or a general-purpose technology may not be particularly important. While AI is undoubtedly a rapidly emerging industry, the real focus should be on how to diffuse it across other sectors, as it seems to represent the general-purpose technology of our time. Would you agree that AI is indeed the GPT of this era?

D: I think for me, I don’t necessarily take a stance on whether it will be or not. I think it’s very hard to forecast which technologies are going to be the most impactful. If we were having this conversation 20 years ago, we would be talking about nanotechnology. But I do think there’s some evidence that suggests AI is becoming a general purpose technology. There have been studies that show the patents that cite machine learning patents, for example, come from a huge diversity of different technological classes, whereas patents that cite blockchain technologies or biotechnology patents might come from a smaller range of technology classes. So that shows that AI is more enabling potentially than some of these other general purpose technologies.

L: So while it may not be 100% certain that AI is the GPT of our time, the possibility seems highly likely. Could you elaborate on the differences between the LS theory and the GPT diffusion theory, and how each explains the rise and fall of economic power?

D: So for the leading sector theory, because it is focused on which country can monopolize innovations in these new, fast growing sectors, the institutional adaptations that it highlights that the countries need to undertake are ones that help countries corner the market in these new industries. So which countries can make sure that their technology doesn’t leak out to other countries? Or how do you attract the best and the brightest? How do you set up the strongest research labs, cutting edge R&D labs? Whereas for the GPT diffusion theory, since the focus is more on which countries can adopt and diffuse GPTs across the entire economy in a decades long, multiple, decades long process, it’s more about which countries can create institutions to train a wide pool of engineering talent associated with the GPT.

Driving Technology Diffusion: The Essential Role of AI Talent

L: The two theories differ in several aspects, including the time frame, the scope and breadth of technology applications, and the corresponding institutional adaptation strategies. Why do you believe the GPT diffusion theory provides a better explanation of the relationship between technological change and the rise and fall of economic power?

D: The book goes through three past industrial revolutions and traces how the new technologies in each of these periods actually influence the rise and fall of economic leadership. I look at the first industrial revolution case where Britain became the clear economic leader. And then I analyzed the US rise in the second industrial revolution, which is the 1870 to 1913 period. And then finally, I look at Japan’s challenge to US Geological leadership in the information revolution in the 1980s. 

In all of these periods, the key explanatory factor was not about which country had innovation leadership, or which country was able to monopolize all the innovations in these leading sectors. I found that the historical evidence provided more support for the GPT diffusion theory. In some cases, like with the US and the second industrial revolution, it was not even that close to being the center of innovation in all of these new industries, but it was more successful at adopting The crucial general purpose technologies of that time, including interchangeable parts, manufacturing, machine tools and electricity at scale.

L: Could you elaborate on the case study of the Third Industrial Revolution? I recall your book includes detailed discussions about the US-Japan competition during that period.

D: What’s really interesting about the US-Japan case is, during that time in the 1960s, 1970s, 1980s, everyone predicted that Japan would eventually become the number one economy and become the technological leader, in part because they were operating based on the assumptions of the leading sector mechanism. They saw Japan controlling key market shares of all of these new industries in consumer electronics, semiconductor components, HDTV and so on. But I find when I go back through this case, the reason why Japan never overtook the US in terms of economic leadership was because Japan was never able to gain an edge in the diffusion of the crucial general purpose technology at the time, which was the computer and this trajectory of computerization. 

L: While your analysis of previous industrial revolutions provides strong evidence for the GPT diffusion theory, I guess those who haven’t read your book might find this perspective counterintuitive in the context of AI. For instance, generative AI only reached maturity about two years ago, yet it has already transformed numerous aspects of daily life and revolutionized countless products and services. This raises the question: is pursuing critical technological innovations in emerging industries truly as unimportant as you suggest?

D: I think the crucial question for generative AI is, how will it actually impact economic growth differentials between countries? Because here, when you talk about generative AI advances, it’s not just the US that has developed these large language models. China has been a fast follower in this space. France, the UAE have all developed open source models that are pretty strong when it comes to generative AI, and then ultimately, the impact comes not from just people using it to play around and have fun. It has to be adopted into business. So I think that will still take a really long road, and a country other than the US can actually become the diffusion leader.

L: So how can we assess a country’s capacity to diffuse general-purpose technologies? In your book, you highlight the importance of skill infrastructure as a key factor for spreading GPTs across industries. Based on your theory, what indicators or factors should we consider when evaluating a country’s strengths and weaknesses in GPT diffusion?

D: So my theory focuses on a small set of institutions that are involved in skill formation. These are education and training systems that widen the base of engineering talent linked to a general purpose technology. And I focus on this because talent and human capital permutes across all other potential institutions. And I think engineering skills play a really important role in adopting general purpose technologies, because that type of engineering knowledge helps standardize and systematize know-how associated with the GPT so it makes it easier for information to flow between the GPT sector and all the different application sectors that are trying to adapt and absorb that GPT according to their own specific local conditions.

L: Are there any other significant institutional factors that you believe are worth mentioning?

D: I think there’s a lot of other factors. There’s been some studies that show a more decentralized approach to science and technology development helps with adapting and diffusing new technologies. Some studies have pointed to a more competitive environment, so less dominance by big business allows for technologies to travel more effectively. But I think I was only able to focus on this, this set of institutions, the set of skill formation.

L: What indicators do you use to evaluate skill infrastructure? Which indicators can help us understand how robust a country’s skill infrastructure is?

D: One of the indicators I use is the number of universities in a country that meet a certain baseline for training average software engineers or average AI engineers. That could be one indicator. Another indicator would be the strength of connections between universities and industry. How strong are those communication linkages that allow ideas to travel from places that are developing new innovations and places that are implementing them? Those are some of the indicators I use.

L: It reminds me of the Taiwan AI Academy. This organization is not a university but a nonprofit dedicated to supporting AI adoption across various industries in Taiwan through the kind of talent cultivation you’ve emphasized. The academy trains professionals from diverse sectors, equipping them with the skills to integrate AI into their industries and address sector-specific challenges. I mention this example because organizations like these play a significant role in expanding the AI talent pool, yet they may not be fully reflected in university-centric indicators.

What’s your perspective on the role of non-university institutions, such as nonprofits and community colleges, in developing AI talent?

D: I think it’s hard to measure those types of institutions, but this is really important for developing the skill infrastructure for AI, because the importance of these non traditional pathways to cultivating this broader talent base, such as institutions like the Taiwan AI Academy, but also, I think you mentioned the community colleges, that provide these alternative pathways. But I think the indicators I used are still useful.

US-China AI Race: What Gives Each Nation Its Edge?

L: Let me zoom out and talk about the US and China. According to your book, the US appears to have a significant advantage in the AI competition with China, which contrasts with many predictions and warnings. Could you share your thoughts on the current state of US-China AI competition and its potential trajectory?

D: I think the US is much better positioned to take advantage of the AI Revolution by adopting and diffusing this GPT throughout the entire economy, and it goes back to some of those indicators we were just talking about. On a lot of these measures, including universities that can train average AI engineers, China only has 29 universities that meet this benchmark I use, whereas the US has 159 of those institutions. So once you get beyond the top tier universities in China, like Tsinghua, Peking, Nanjing University, China just doesn’t have as deep of a bench, deep of a reservoir of institutions that can cultivate AI engineering talent to adopt AI throughout all these different economic sectors. And the US has better linkages between industry and academia to transfer technologies across different parts of the science and technology system.

L: Could you elaborate on your views regarding the current tech policies of these two countries? What are their main focuses, and how might these policies enhance or undermine their economic competitiveness? Also how do you think the outcome of the US elections might shape the country’s future AI policy?

D: So first, I think both the US and China are preoccupied with the leading sector approach to technological competition. For the US part, a lot of the current administration’s policies are obsessed with preventing the leakage of crown jewels in AI, adopting this sort of fortress America style approach. China as well, is very focused on treating AI as a strategic sector and focused on self sufficiency in AI, encouraging technological self reliance. And China, for its part, seems very focused on cutting edge R&D in AI, as opposed to this broad based education and diffusion oriented approach. 

So my recommendation for both countries, or rather any country, is to adopt a more diffusion centered approach to competing in AI. Regarding the election results in the US, I think one interesting quirk is that there seems to be a bipartisan consensus that the leading sector approach is the right way to compete with China in AI, so in that respect, I don’t know if it would make a meaningful difference.

L: That’s a very insightful observation. But now both the US and Taiwan place significant emphasis on regulations like investment screening and export controls to prevent critical technologies from falling into the hands of competitors. And it seems you’re not entirely convinced by this approach. Could you share more of your thoughts regarding the necessity and effectiveness of such containment strategies? Do you think these policies have a role to play in helping the US or Taiwan maintain their competitive edge in AI diffusion?

D: I think these approaches, the technology controls, are more appropriate for certain types of technologies and less appropriate for others. I think they’re not very well suited for dealing with general purpose technologies, because no one country can monopolize technological advances in a field as broad and as fast moving as a GPT. It would be like trying to say we’re going to bottle up electricity within the US or within Taiwan. 

I think the other reason why the technology control mindset is not as suitable for handling GPTs is that the control approach might be effective for maintaining a lead in the next two to three years, but for general purpose technologies, my book argues they don’t make their mark until after multiple decades of diffusion, and by that time, it’s very hard for one country to say we have the only technology that matters in this area. So those are the two reasons why I think technology control approaches are ineffective.

L: So it seems you don’t view these regulatory tools as particularly necessary. Are you suggesting that we might not need export controls and should instead focus on developing AI talent, rather than restricting the export of advanced chips to prevent them from reaching competitor nations?

D: I think part of it is, if it’s ineffective, then it’s not going to bring many benefits, and it’s only going to create disadvantages. So for example, the US October 2022 export controls on high end chips hurt our strongest chip company, Nvidia, and also escalated to a type of economic containment approach towards China, which could increase tensions between the two countries. So if there’s limited advantages and only these drawbacks, then I don’t really see the logic behind the policy.

L: Simply put, you suggest that we should prioritize building long-term advantages through technology diffusion rather than focusing solely on regulating the outflow of specific technologies. 

D: Yeah, I think for whether it’s the US, China, Taiwan, or any other country, my book’s recommendations are adopt a diffusion centered strategy to upgrade the overall AI engineering talent in your country, improve the connections and technology transfer institutions and linkages between different parts of the science and technology ecosystem.

More than a Silicon Island: Charting Taiwan’s AI Strategy for the Future

L: Do these recommendations also apply to middle powers like Taiwan, or would you have different suggestions?

D: I think one of the considerations for a middle power, or a middle income or newly industrializing economies, is once you have the established absorptive capacity to be connected to the technological frontier, then you can focus on adoption and diffusion. I think some countries don’t have that requisite absorptive capacity, like say, a country like Venezuela does not have frontier firms connected to the AI space or frontier universities connected to the technological frontier in AI, so they’re not able to absorb advances incubated from abroad. Taiwan is embedded in those networks. Taiwan is the home base to a lot of foreign multinational corporations, research bases and talent bases. Taiwan has strong universities and firms connected to the latest advances in AI and so they can pursue a diffusion centered AI strategy. 

L: Are you saying that Taiwan is well-positioned not only to excel in areas like AI chip manufacturing but also to demonstrate strengths in AI applications and cross-sector diffusion, provided we continue to enhance our ability to spread AI technologies through the right policies?

D: I think Taiwan can view technological leadership, not just in terms of single sector dominance, but more about the capacity to broadly diffuse foundational technologies like AI across all different sectors of its economy. It’s harder to pursue that type of strategy because you can’t just focus on one industry. You have to focus on long term investments in things that affect the entire population. But it’s still possible.

L: Based on what you’ve mentioned, one of Taiwan’s key advantages in promoting AI technology diffusion is its integration into global AI development networks. My final question relates to this: how can countries collaborate to enhance the diffusion of AI technologies? Particularly in the context of global economic competition, there’s increasing emphasis on strengthening alliances among democratic nations. In your view, how can like-minded countries cooperate to gain a greater edge in the AI race?

D: Taiwan can benefit from these international collaborations by ensuring that the time from a new breakthrough in AI to a Taiwanese frontier firm or university implementing that breakthrough shrinks. And so that’s an example where Taiwan can get in, continue to further embed itself in these global innovation networks and AI. 

But I think really, for me, that’s not necessarily the crucial gap. That gap is already shrinking, with or without government policy, just with globalization. The crucial gap is the intensive adoption gap between once a frontier firm in Taiwan applies a new AI application for the first time, then how much time does it take to spread throughout the entire country? That’s something international alliance doesn’t necessarily affect. It’s going to be more about Taiwan’s domestic ecosystem.

L: Thank you for sharing your insights. I truly appreciate your time today and look forward to the release of the Chinese edition of your new book!

D: Thank you as well for preparing such thoughtful questions for this discussion!