Is China Taking the Lead in AI?

In 2017, the Chinese government announced plans to “lead the world” in artificial intelligence by 2030. The announcement has fed considerable uneasiness in the United States and elsewhere about the scope of China’s aspirations and the extent to which the country might use AI to tighten control over its citizens and develop more sophisticated military capabilities. However, the anxiety over China’s plans for AI may be overblown, says Jeffrey Ding, an economics and technology researcher at the University of Oxford’s Future of Humanity Institute. Yes, there are many signs that China is making huge investments in AI, and it leads the world in AI-related patent filings and publications — in 2017 alone, it won some 900 patents related to facial recognition, compared with fewer than 150 in the United States. However, Ding says, China’s overall position is more complicated and requires a nuanced view.

Ding, who shared his perspective last year with a U.S. congressional commission that monitors national security and trade issues between the United States and China, is author of a white paper titled “Deciphering China’s AI Dream” and publishes a newsletter on AI in China called ChinAI. MIT Sloan Management Review correspondent Frieda Klotz spoke with Ding about the current state of AI in China and how he expects the future to unfold. This is an edited version of their conversation.

MIT Sloan Management Review: Many people have suggested that we’re in the midst of an arms race between the United States and China with regard to AI. Are they right?

Jeffrey Ding: The arms race metaphor persists because AI is such a strategic technology. It’s easy to plug it into the old narrative of U.S.-Soviet competition over other strategic weapons, such as nuclear capacity, during the Cold War. Yes, there are some similarities in the sense that two powers are competing over technology, but the analogy is flawed. For one thing, the technology is different — AI is a general purpose technology that applies across many domains. A lot of the benefits will be economic rather than specific to a single field like defense.

The arms race meme is dangerous because it creates a sense of zero-sum competition. We’re talking about something very broad — AI incorporates natural language processing, computer vision, predictive analytics, and many other domains, all of which have lots of different applications. So you need to be specific about which domain you’re talking about. Claiming China has an edge is an oversimplification.

What do you see as the main myths about China’s approach to AI? What are people in the United States and other countries most worried about — and what should they be worried about?

Ding: One big misconception is that China’s strategic thinking about AI started with its 2017 plan. The Chinese government had already developed plans for related areas, such as the internet of things, smart manufacturing, and science and technology. On top of this, local governments and private companies had been making substantial investments in the AI space. So, in a sense, the central government is just following in their wake.

A second myth is that China is taking a monolithic approach to AI. In reality, different local and regional governments are taking different types of steps — for instance, giving subsidies to specific companies. It’s not the highly organized, top-down strategy people might think it is.

Perhaps the biggest myth is that China’s AI capabilities are enormous. Part of this derives from what I call the AI abstraction problem, where people cherry-pick examples of where China might be ahead — such as in facial recognition or smart surveillance — and assume that applies to other areas. However, the fields in which China has big leads are not areas that are likely to spread benefits across the entire economy. Based on my analysis, China is lagging behind the United States in every metric except access to data.

Can you be specific? How can we assess China’s potential in AI?

Ding: It is very important to be specific. I have looked at indicators such as hardware, data use, research, and commercial applications of the technology. China occupies 4% of the world’s hardware market, while the U.S. is at 50%. In the commercial sector, there are twice as many AI startups in the U.S. as in China.

Although some observers like to point to who has more patents and more publications, I think a better metric is who is able to apply innovations faster and at a wider scale. In the U.S., people can study machine learning and data science outside of formal educational institutions, in places like coworking spaces. This provides opportunities for creative applications of new technologies. In China, universities are developing their own AI colleges. It’s still not clear which strategies for diffusing technologies into the mainstream will be most effective, and which country will be most successful in adapting its institutions so that they can take advantage of AI.

What are the fundamental differences between how China and the U.S. are approaching AI?

Ding: To start with, you have to recognize that China and the United States are in very different economic positions. China is trying to escape the middle-income trap and is hoping AI can help it boost its manufacturing outputs. By contrast, the United States has a more service-oriented economy, and AI may be the key to improvements in industries like finance and enterprise — enhancing back-end logistics, communications, transportation, and forecasting. So rather than saying that certain countries are ahead in AI generally, it’s better to think in terms of what makes most sense for the particular environment.

Where do you see China having a competitive advantage?

Ding: One area that’s drawing a lot of attention from companies is natural language processing (NLP), which is a subdomain of AI designed to provide people with better chat-conversation assistance. Chinese companies have a substantial portion of the NLP market that uses Mandarin, while American companies have a better hold on English.

Another area where China is pushing ahead and is eagerly diffusing the technology is facial recognition, especially with regard to surveillance and security applications. In 2017, China received about six times more patents in this area than U.S. companies.

As you know, facial recognition has generated a huge amount of concern based on the fact that China has deployed it to track minorities and suppress dissent. Are there applications that are less highly charged?

Ding: Facial recognition has a number of applications in smart security. It can be used for identity authentication — for example, on mobile phones. But even then, there can be security problems. What’s more, the technology doesn’t seem to have that many productivity spillovers to other industries. It’s not clear that being the leader in that area will yield broader economic benefits for China.

How do you see AI strength playing out for the United States and other countries?

Ding: The real economic advantages will come from improvements in infrastructure — things like smart grids that deliver electricity when it’s needed — and product advances like autonomous vehicles, [such as] the self-driving cars Google and Tesla are developing. The United States and Europe are leading in the technologies that feed these areas, like AI open-source software and hardware. China lags far behind.

Google, as you know, recently got into trouble for accessing health care data without permission. It’s widely assumed that Chinese companies are less inhibited by privacy concerns. Is this true, and if so, do you think it will enable China to take the lead in health care?

Ding: In China, it’s true that researchers can collect more data on hospital patients from a single hospital, whereas U.S. privacy laws require researchers to work with data from different hospitals after a certain threshold has been reached. So this is sometimes seen as an advantage for China, but I think it’s overstated. First, health care data sets are not really national assets. Typically, they are used jointly by researchers from different hospitals and universities; U.S. researchers might have as much access to the data as the Chinese hospital where it came from. There are many collaborative research ventures and alliances where data is shared across national boundaries — or if it isn’t shared, researchers have access or can derive benefits from it.

To deploy AI in health care effectively, you need standardized systems and electronic health records. China got into this relatively late — its electronic health record systems still require a lot of public investment to catch up. We’re still in the early innings of determining who is diffusing AI technology faster or more sustainably in health care.

One more myth I haven’t yet mentioned relates to health care and data privacy: the idea that China doesn’t care about AI ethics. In fact, there’s evidence that academics, bloggers, and Chinese citizens are paying more attention to privacy and other concerns. Now, it’s true that some issues — such as the disproportionate targeting of ethnic minorities for surveillance in Xinjiang — are considered off-limits. But these are notable exceptions. I’ve translated blogs by Hu Yong, a professor at Peking University, who complains that the government infringed on citizens’ privacy in the public health battle against the coronavirus. Discussions are definitely taking place. The big difference is that the Chinese government works hard to suppress civil society, making these kinds of discussions less visible and harder to find.

You presented your views on China’s AI capabilities in Washington last year. What has changed in the meantime?

Ding: The overall picture I painted last year, with the U.S. dominating in AI practice and expertise, remains broadly the same. In my view, the United States, with its universities and open innovation ecosystem, has important structural advantages that enable it to attract the best and brightest. These advantages may even enable it to extend its lead.