How might China win the future? Ask Google’s AI
The country’s sprawling industrial policy is beyond mere human comprehension
If China dominates the 21st-century economy, its industrial policy will get a lot of the credit. The state’s efforts to cultivate new industries, breed winners and foster technological advances inspire awe and anger from outside observers. Kyle Chan of Princeton University recently compared China’s policies to the Manhattan Project, which invented the atomic bomb. On present trends, he argues, “the battle for supremacy” in artificial intelligence (AI) will be fought not between America and China but between leading Chinese cities like Hangzhou and Shenzhen.
American officials may scoff—but they do not doubt the power of China’s policy to distort markets. In March America’s trade representative released a report criticising the “guidance, resources and regulatory support” that China’s government bestows on favoured industries, at the expense of foreign rivals. The report called the “Made in China 2025” plan, which increased the country’s share of industries like drones and electric vehicles, “far-reaching and harmful”. Such grievances help explain why President Donald Trump hit China with punishing tariffs in April.
One problem facing both critics and students of China’s industrial policies is that it has so many. Its efforts are not confined to national plans such as Made in China 2025 or discrete undertakings akin to the Manhattan Project. They also take the form of obscure memoranda, lacking pithy names, issued by local officials—policies like Anqing City’s “Notice on Supporting Large Commercial Circulation Enterprises In Carrying Out Pilot Projects For The Construction Of Modern Rural Circulation Systems”.
China churns out more than 100,000 policy documents a year. Over a fifth feature some kind of industrial policy, according to a new paper by Hanming Fang of the University of Pennsylvania, along with Ming Li and Guangli Lu of the Chinese University of Hong Kong, Shenzen. Not even the most diligent researcher could wrap their head around them all. Fortunately, one of the emerging industries that China is keen to promote—AI—can now help economists understand how China goes about its task.
Mr Fang and his co-authors used Google’s Gemini, a large language model, to scour millions of documents issued from 2000 to 2022. They were scraped from official websites or obtained from PKUlaw.com, a vast repository. When properly prompted, Gemini could identify documents that met the definition of industrial policy. It could also discern the industries targeted and the tools used. To make sure the model was not making things up, the authors took precautions. They told the model to think of itself as an expert in Chinese industrial policy, and carried out random spot checks themselves. They then asked another model (from OpenAI) to refine the findings. The result is a rich database covering two decades’ worth of Chinese efforts to win the future.
Their work captures the underappreciated variety of China’s tools. Subsidies are most popular, appearing in 41% of policies. But they are only one tool out of 20 and often absent. Cheap credit and land are less common than you might expect, featuring in less than 15% of documents. Protection from foreign competition appears in only 9%. Instead of offering handouts, protection or perks, about 40% of central-government policies regulate the targeted industry by, say, imposing quality or efficiency standards. A few policies—3%—try to suppress industries that might be too dirty, inefficient or otherwise undesirable. Equating industrial policy with subsidies and tariffs, then, will “paint an incomplete picture of China’s industrial-policy landscape”, the authors warn.
That landscape has also changed over time. Tax breaks and explicit protectionism have become less fashionable. Government funds, which are supposed to act like venture capitalists, have become more so. Efforts to cultivate supply chains and clusters have also become more widely used. The same is true of policies to increase demand, such as government purchases or consumer subsidies. They appear in about a fifth of policy documents, roughly double their share in 2000. Often, city governments embrace these new tools before provincial or central governments.
One of the biggest surprises is that manufacturing—making in China—is targeted by only 29% of the policies. More focus on services; 17% are still dedicated to agriculture. Within these broad sectors, targeting can be savvy. City governments are more likely to pick industries that loom large locally, relative to their prominence province-wide, suggesting officials are playing to their city’s strengths. This is particularly true in richer cities, where local governments may be more sophisticated.
Some cold water
Targets are, however, becoming more similar. All 41 cities in the Yangzi River Delta have chosen “high-end manufacturing” as a “key” industry in their five-year plan, notes The Paper, a Chinese website. Thirty-eight have picked new-generation information technology, which includes AI. The delta, the paper worries, suffers from “industrial isomorphism”.
Local governments may be converging on similar approaches because they are all trying to follow the central government’s lead. Cities have become more likely to cite national policies in their own documents since 2013. They may also be emulating each other. Unfortunately, policies seem to become less effective as they spread from pioneers to laggards. Measures that lift firm revenues and profits in leading cities may fail to replicate in latecomers. Not everyone can have their own Manhattan Project.
Overall, the link between industrial policy and the productivity of firms is “mixed and tenuous” the authors say. But with the assistance of Gemini, they have shed some valuable new light on industrial policy’s many variations. Maybe AI can help improve the design of policies to encourage emerging industries like AI. ■