Jeffrey Ding on Examining AI Development Between the United States and China

Jeffrey Ding is the China lead for the Centre for the Governance of AI. Jeff researches China’s development of AI at the Future of Humanity Institute, University of Oxford. His work has been cited in the Washington Post, South China Morning Post, MIT Technology Review, Bloomberg News, Quartz, and other outlets. A fluent Mandarin speaker, he has worked at the U.S. Department of State and the Hong Kong Legislative Council. He is also reading for a D.Phil. in International Relations as a Rhodes Scholar at the University of Oxford.
Nohl Patterson CMC '22 interviewed Jeffrey Ding on September 14, 2020.
Photograph and biography courtesy of Jeffrey Ding.

China has reiterated its dream to become a technological superpower. China has publicly stated they are going to double their STEM Phds by 2030 and poured money into primary and secondary artificial intelligence (AI) research. What do you think China’s long term objectives are?

I do not think this a dramatic new approach, it's a continuation of rhetoric we saw back with Deng Xiaoping and this emphasis on science and technology as primary productive forces. That has continued through different shifts in leadership. Actually, the investment of $150 billion into primary AI research is a common misconception of what the AI plan outlines. That figure is a benchmark for the goal of the total output for the AI industry, not how much they have put into AI investments. This is a common misconception because a lot of times in English-language media coverage and analysis, there is a tendency to overestimate China’s AI capabilities. That figure pops up a lot of times, it is not that they are pouring in that much money but that this is their benchmark goal for the entire industry's output by 2030. I think a couple of organizations, the Institute of Defense Analysis and Center for Emerging Technology, have done separate comparisons of US and China research and development (R&D) spending and most of the conclusions find that there is too much uncertainty to decide which side is spending more on AI.

I am curious why do you think that Western news organizations have a tendency to over rotate on this issue and dramatize China’s AI?

There are many reasons, but one reason is that it is a difficult technical question to clearly define what counts as AI. It is such a malleable term if you want something to be AI it can be AI. It has become this magic buzzword. There is this malleability associated with it. For example, if you are someone in the military bureaucracy and want to push for more funding for your programs involving AI, you have an incentive to overestimate the threat from China’s AI advances, to gain more momentum for your program. Interest groups have a lot of reasons to inflate threats, but I think it is a part of this larger dynamic where you are seeing broader structural pressures towards US-China competition, especially as China continues to grow. You have seen these overestimates before with other countries' technology- the missile gap during the Cold War era and with Japanese tech during the 1980’s.

How does a country foster innovation and drive technology growth? While this seems like an abstract process, there is some well-documented evidence that different approaches yield different results. Do you think that China’s approach to AI is best categorized as a top down approach and if so, do you think this an effective long run strategy?

Yea, I think there is definitely more direct government intervention targeted at specific industrial sectors. It definitely applies to AI and to related fields such as semiconductors. That is partly due to the fact that countries are catching up to the technological frontier, where the U.S. sits. For those countries there is a clearer roadmap of where to follow to catch-up. For countries at the frontier, like the U.S., it just makes more sense to have a more decentralized approach, because it's more likely that it will produce the next technological innovation. At the same time, I don't think that China's approach to AI policy is as top-down as in other technology domains like semiconductors, high-speed rail, or solar panel manufacturing. This is because there was a lot of pressure from local governments to put money into AI. Provincial government really led the charge and the central government largely followed in their wake.

In your paper, “Deciphering China’s AI Dream” you spoke to a similar topic. Can you talk through your thesis about local and provincial governments being a driving force behind innovation and AI?

In general, the central government laid out a broad signal that AI was a national strategic priority, but the implementation of these science and technology policies happens on the ground level. There have been a few local clusters for AI policy and AI development. These clusters are beyond the big four technology ecosystems that are already developed and mature: Beijing, Shanghai, Guangzhou, and Chengdu. Those first-tier cities already have a mature technology ecosystem, so it is natural for AI to develop. However, in other areas like Hangzhou and Hefei, you have had local government partnering with anchor universities and companies. In Hefei, the local government is trying to build China’s Speech Valley- a cluster focused on building better natural language processing capabilities (NLP). Hefei has partnered with IFI Tech, an NLP tech company, and University of Science and Technology to help startups build better NLP capabilities. In Hangzhou a similar phenomenon with the city government trying to set up an AI town. They have partnered with Alibaba and Zhejiang University as the key anchor tenants to help startups grow and ideas to flow. For example, in Hangzhou’s AI town, that is a local government initiative housed within Hangzhou’s broader science and technology innovation park. You get geographic proximity with dedicated office space for companies and startups that want to move in. That office space is subsidized or given tax breaks. The funding support also comes through companies applying for grants for cloud computing resources provided by Alibaba along with R&D subsidies and tax breaks. 

In “Deciphering China’s AI Dream” you talk about China’s protectionist technology nationalism. Do you think in the long-run this strategy is a recipe for success and what are the short-term consequences for this approach?

China has to strike a delicate balance between technology nationalism and technology globalism. On one hand it is almost impossible to maintain scientific and technological power without being plugged into the global innovation network. For this reason, complete protectionism is simply not feasible. On the other hand, China is very wary of dependency on rivals and adversaries. For example, the cutoff of semiconductor chips is probably the most salient concern. They are trying to balance staying plugged into global innovation networks and not becoming overly dependent on a single source from those networks. The large data China has is often seen as a key strategic advantage for China. However, I do not know if this is pivotal or even a strategic advantage because there is so much variance in the domain we are talking about. For mobile applications there is some advantage because of the mobile internet’s scale in China. But, if you are talking about AI as a general-purpose technology, any AI application will spread across these different domains. When you talk about a “data advantage” it has to be domain specific. For instance, having more mobile users does not help autonomous vehicle development. All of this is also dependent on trends in the technological trajectory of AI. Currently in the field there are more and more advances towards getting good results on smaller samples of data and advances of synthetic data- generating one’s own data. These innovations largely nullify the advantages associated with having more data.

What do you see as China’s biggest hurdle to overcome, technical or political, in its quest for achieving status as a technological superpower?

One area where it's very hard for technological capabilities to diffuse is in complex systems. Gilli and Gilli wrote a piece in International Security in 2019 about why China will not catch up. They highlighted several advanced weapon systems that require an immense amount of systems integration capabilities. In the U.S., prime contractors have to organize thousands of suppliers and components to make something like a stealth fighter. Those capabilities are very hard to steal, imitate, or diffuse to other leading militaries. For myself, I see the biggest challenge as not a technological innovation or research advancements problem, which is often held up as one of the biggest barriers. I believe the biggest obstacle for China is not having strong diffusion capabilities. This means the ability to take advances from a leading firm in Beijing has already been adopted, such as a smart city solution- technology that manages energy or solves traffic problems.  Taking that new innovation and spreading it across China into a bunch of different contexts is the biggest challenge. It plays into all these other problems such as rural and urban divide, income inequality- how do you get this technology from first tier cities to the rest of China?

There are some unique aspects of AI development that make it different than developing more conventional technology. One comparison to contemporary efforts to develop more advanced AI has been the U.S. developing the hydrogen bomb by recruiting the foremost physicists in the world during the 1930’s and 1940’s. It has been cited that the U.S. possesses a significant majority of the top talent in AI. Does having the top talent in one spot give the United States a competitive advantage for developing better AI or even a general intelligence AI?

For more breakthrough advances in AI, like gameplay and uncertain decision-making, the record shows that those advances come from a cluster of top talent such as Carnegie Mellon University, Open AI, and DeepMind. I do think that while you can downplay the importance of fundamental research, to make those advances you need that top talent to innovate. This goes back to the question of whether you need the top talent or can you get away with a lot of people working with the data, even if it is just a “plug and play” algorithm approach. For the U.S. the academic job market for that top talent is tightening up so a top priority is connecting the top Phd talent with industry. The most salient effects will not necessarily come from more STEM Phds, but getting more technically trained people to work on the engineering work to deploy the capabilities.

Looking forward, what is the most significant hurdle the United States will face in the next twenty years for AI development?

I think that one of the toughest challenges is going to be job displacement. The speed and scale of automation and its effects on employment rates. There has been some recent research that compared to past waves of automation, the current wave is displacing more jobs than it is creating. This could be a worrying trend for the future. An associated risk is that people who are less technically inclined will be against the development of the technology displacing them. It gets at a deeper question “What kind of society do we want to build?” One of the worries I have is that as algorithms get more precise and more personalized, I think there will be even more risk for addiction spirals. It will be one of the tough issues we have to grapple with. There is an inexorable push towards more productivity and AI will certainly bring that, but the issue is “at what cost?”

Nohl Patterson CMC '22Student Journalist


Share this:

Leave a Reply

Your email address will not be published. Required fields are marked *