Economist Dieter Ernst agrees there is a lot of AI-related activity in China, but argues it's the wrong kind of activity, not likely to result in much -- for now...
Who’s winning the U.S. vs. China AI battle? It’s a question frequently asked but too often poorly answered. The responses tend to be either oversimplified or exceedingly complex. The messages could be muddled by both sides, as the United States and China try to impose their own, politically motivated answers, colored by the trade war, and now further complicated by acrimony over the coronavirus outbreak.
In the United States, China is widely believed to be building a technological lead in artificial intelligence. That belief is challenged, however, in a recently published study on China’s AI chips, written by economist Dieter Ernst. His 70-page report provides multi-layered answers with nuances textured by Ernst’s social, historical and industrial knowledge of China.
Ernst, China expert and keen observer of the global semiconductor industry, is senior fellow at the Centre for International Governance Innovation (CIGI), which is based in Waterloo, Canada, and the East-West Center, based in Honolulu. CIGI, an independent, non-partisan think tank, published Ernst’s peer-reviewed research and analysis entitled “Competing in Artificial Intelligence Chips: China’s Challenge Amid Technology War.”
In his report, Ernst explains his objectives: to “fact check the assumptions underpinning America’s claim that China is about to forge ahead as an AI technology leader.”
Using field studies conducted in China in 2019, Ernst goes to the heart of the matter: What’s happening in China’s AI chip industry?
Companies covered in the report range from Huawei/HiSilicon, to digital platform players like Baidu and Alibaba, to China’s AI chip unicorns, including DeePhi, Cambricon, Horizon Robotics, Yuntian Lifei/Intellivision, and Suiyuan Technology.
Getting to know what each of these players is doing is immensely valuable. But even more helpful is the way Ernst walks readers through why and how China’s AI chip industry has arrived at where it is today.
The industrial and historical background that Ernst provides is crucial to understand the big challenges for China’s AI ecosystem. The report also cautions that China’s AI development trajectory could be altered if the United States, driven by “crude techno-nationalism,” continues to cut off China’s access to US technologies.
Ernst concludes that fears about China credibly threatening America’s leadership are “not grounded in reality.” Rather, he believes this position reflects “a general China-bashing mode in Washington, D.C., that is driven primarily by ideology and geopolitical considerations.”
Here are the key findings Ernst discusses in his report.
Where Ernst’s research stands out from other reports is his focus on “the nexus between AI research and industry” in China.
A common narrative on China, prevalent in the United States, often says that China’s top-down, monolithic innovation policy has led to economic success. The report calls this a “myth.” Ernst argues that things are never so simple nor so homogeneous in China. The report quotes Harvard’s Mark Wu: “What makes China complicated is that, while the Party-state holds vast control levers, it allows market forces to play out in huge swaths of the economy.”
The intertwined nature among the state, the Party, state-owned enterprises, private companies, financial institutions and others is not actually helping China to accelerate AI development. Rather, it is symptomatic of “the surprisingly fragmented Chinese innovation system,” Ernst noted.
Ernst also discusses how China’s innovation system is “constrained by multiple disconnects between research institutes and universities on the one hand and industry on the other.” The disconnects include friction between the state seeking to dictate top-down technological choices, and local governments chasing hot opportunities.
Consider, for example, “Made in China 2025.” Many Western observers believe China’s recent policy initiatives have become an engine fostering innovation. Ernst doesn’t agree. He is more concerned that that this sort of centralized control could only “slow down China’s transition to a more market-oriented approach to innovation policy.”
Ernst observed in his report:
There is no doubt that China will continue to generate a significant flow of AI-related innovations. Yet institutional structures surrounding Chinese innovative efforts are likely to create large-scale misallocation, waste and resource leakage. The resultant disconnect between AI research and industry is real, immediate and unlikely to fade away any time soon.
Three trajectories of China’s AI development
In explaining China’s AI development, Ernst divides the current situation into three separate trajectories.
a) Public research organizations
The first trajectory is mainly driven by state-related institutions such as public research organizations, universities, and powerful state-owned enterprises conducting AI research in line with objectives outlined in China’s “New Generation Artificial Intelligence Development Plan” (AIDP).
One concern raised by Ernst in the report, however, is that “there is little systematic attempt to ensure that industry is exposed to the research findings included in those [academic papers] publications and the patents.”
Judging from responses his team received in interviews with Chinese AI companies, Ernst observes that “knowledge exchange with public research institutes has only very recently gained in importance.”
As for the first AI development trajectory in China, Ernst concludes:
…both for AI publications and for AI patents, interactions between public research institutes and industry remain limited — “a lot of AI knowledge is lying idle at universities and research institutions”
b) Digital platform leaders
He cites China’s digital platform leaders including Huawei, Alibaba, Tencent, Baidu and Lenovo as key movers of the second trajectory of AI development.
Given the scale and the resources of these companies, they appear to hold the key to the future of China’s AI R&D.
Ernst’s interviews, however, reveal that for these national leaders, “investment in AI applications was the top priority.” He noted, “By contrast, investments in the development of algorithms and AI chips were considered to be ‘too little’ and ‘insufficient.’”
Curiously, Ernst’s team was told that “close interaction with public AI research has only gained in importance since the outbreak of the [US-China] technology war.” The escalating trade/technology battle is now forcing Chinese companies to innovate in areas formerly considered a “US comparative advantage.”
c) AI unicorns
China’s AI unicorns are a driving force behind the third trajectory of AI development. Ernst divided these numerous companies into two categories.
The first group consists of those using “existing machine-learning algorithms, primarily neural networks, to sell AI application software.” The second group are typically involved in the design of AI chips. They receive substantial support from the government, reports Ernst.
Ernst’s interviews in China reveal that “many of the first group of AI unicorns can hardly keep up with the rapid demand growth for AI applications in the Chinese market.”
To move ahead quickly in application markets, “these companies are trying to recruit as many young engineering graduates as they can from across China. They also are fiercely competing for experienced top talent from overseas.”
As often discussed, most Chinese AI startups do not invest nearly enough in cutting edge technology. Ernst shares insights about the peculiar nature of China’s stock exchange. Unlike the United States, China’s exchange requires companies to be profitable for at least three years before going for an IPO. This practice appears to penalize AI startups in China with heavy R&D spending. It prompts many to look for an easier way out — focusing on AI applications.
While China’s digital platform vendors have shown an appetite for acquiring AI startups (exemplified in Tencent’s funding in Suiyuan Technology, and Alibaba’s acquisition of C-Sky Microsystems, notable investment arms — one created by Huawei and another by Xiaomi), Ernst concludes that the results of these acquisitions and investments “remain unclear.” He wrote: “It will take some time to separate the wheat from the chaff.”
Did Lee Kai-Fu get it right?
Ernst’s position is that China’s AI development remains fragmented, and yet there is demonstrably a great deal of AI activity in China. How to explain these two observations?
Ernst starts by examining the comments of one of the leading voices arguing that China is building an AI lead. He cites Lee Kai-Fu, a former Apple, Microsoft and Google executive who is now a venture capitalist. In Lee’s analysis, published in 2018, he wrote:
China can outcompete the United States because AI has moved “from the age of discovery to the age of implementation, and from the age of expertise to the age of data.” For China, what matters now is “the power of data.”
Lee says the key to China’s success is in “China’s huge population of relatively low-cost college graduates who will toil away for long hours to do the repetitive work of categorizing huge troves of data needed to ‘train’ AI algorithms.” He suggests that China should use its “big data treasure trove” to get ahead in mass markets for lower-cost AI applications.
While much of China’s AI industry followed Lee’s advice, Ernst believes this strategy didn’t turn out so well for China.
Ernst argues that Lee’s proposition is based on a “fundamental misreading of AI’s permanent research revolution.” This is because two phases in AI development — the age of discovery in AI and the age of implementation in AI — are happening in parallel. Given that AI is a field still evolving while a host of new types of neural networks are emerging, China can’t just live on AI applications based on outdated neural networks.
Perhaps, one of the most important messages Ernst conveys is this:
Our research finds that US technology restrictions are forcing China to strengthen basic and applied AI research to catch up in core foundational technologies.
His team’s interviews in China indicate that “interactions between China’s three AI development trajectories may be beginning to change. Ironically, US technology export restrictions are thus forcing a reform of China’s technological investment and innovation policy, which may help China to correct one of the fundamental weaknesses of its innovation system in AI.”
In EE Times’ interview with Ernst conducted via email, we asked about implications for U.S. industry. He wrote back:
It is time to accept that the US, the most powerful country in the world, can no longer single-handedly dictate the pace of innovation in AI and in IT at large. It is in the interest of US industry that the US government returns to a policy that promotes rather than disrupts the rule of law in international trade, in order to regain stability, predictability and a more equitable distribution of gains from trade. It will take quite some time to repair the tremendous damage done by current US policies.