“SOUT OF THE Few Huai River geese can be seen through the rain and snow. In classical Chinese, this verse is a breakthrough, not in literature but in computing power. The line, composed by an artificial intelligence (AI) language model called Wu Dao 2.0, is indistinguishable in meter and tone from ancient poetry. The lab that built the software, the Beijing Artificial Intelligence Academy (BAAI), challenges visitors to his website to distinguish between Wu Dao and the flesh-and-blood masters of the eighth century. Anecdotal evidence suggests it fools most testers.
The system, whose name means “enlightenment” and which can emulate lower types of speech, derives its power from a neural network with 1.75 trn of variables and other inputs. Google Tags-3, a similar model built a year earlier by a team of researchers in San Francisco and deemed impressive at the time, considered only 175 billion parameters. As such, Wu Dao represents a leap in this type of machine learning, which attempts to mimic how the human brain works. This delights lovers of classic literature, but not as much as the communist authorities in Beijing, who put AI at the heart of China’s technological and economic master plan defined in 2017. It frightens Western governments, who are worried AI‘s less benign applications in areas such as surveillance and warfare. And it intrigues investors, who are spying on a huge business opportunity.
At first glance, the plan is off to a good start. The logistics branch of J.D..com, an e-commerce group, operates one of the world’s most advanced automated warehouses near Shanghai. In May, Chinese search giant Baidu launched driverless taxis in Beijing. SenseTime’s “smart city” AI Models of city surveillance cameras that track everything from traffic accidents to illegally parked cars have been deployed in more than 100 cities in China and overseas. China is deploying more AIassisted industrial robots than any other country. And in 2020, it overtook America in terms of journal citations in the field.
The five most rated Chinese AI specialists are collectively worth nearly $120 billion (see Chart 1). The largest of these, Hikvision, has a market value of $60 billion. SenseTime, which went public in Hong Kong on December 30, is worth $28 billion. Two more are expected to register soon. In 2020 investments in the unlisted AI startups hit $10 billion, according to the AI Index compiled by researchers at Stanford University. In its prospectus, SenseTime anticipates that revenues from AI-assisted image recognition and computer vision software, the most mature part of the market, could reach 100 billion yuan ($16 billion) by 2025, from 24 billion yuan in 2021 (see chart 2).
Look past headlines or fancy Wu Dao verse, however, and things seem more complicated. Yes, China has made progress on AI, and even the occasional big splash like Wu Dao. But it almost certainly still lags behind America in terms of cutting-edge investment and innovation. In 2020, three years after the start of the master plan, private Chinese companies AI companies received less than half of the investment than their American counterparts. And much of the public and private money poured into the sector could end up being wasted.
Five-year-old Chinese child AI master plan sets out a number of objectives. For example, by 2025, the country must create an industry with global revenues of 400 billion yuan, achieve “major breakthroughs” in technology, and lead the world in certain applications. Five years later, it was to dominate the industry (then worth $1 billion in sales), having written its code of ethics and set its technical standards, just as Europe and America defined the contours of the revolution. industrial.
The elements of the Communist Party’s approach are typically normative. The Ministry of Science and Technology has instructed Chinese tech giants of existing enterprises in certain sub-disciplines to AI—Tencent in medical image recognition, Baidu in autonomous driving — to overtake them. That said, the plan is less practical than some of the country’s other development projects, observes Jay Huang of Bernstein, an investment firm. According to Oxford University’s Huw Roberts and five co-authors, the plan primarily acts as a “stamp of approval” that “derisks” assorted AI initiatives carried out by State entities, local authorities and the private sector.
In practice, risk reduction consists of distributing a lot of public money. Some of this comes in the form of tax breaks and grants, such as in the “little giants” program to support 10,000 promising startups in various sectors, including AI. Local governments, even in impoverished Rust Belt provinces like Liaoning in the far northeast, have also dangled similar incentives before AI-curious companies.
Another type of support comes from public procurement. Companies do not disclose the amount of revenue they derive from public sector contracts. But the share is likely to be significant. Central and local authorities use SenseTime’s surveillance technology. Megvii, also specialized in image recognition, maintains many relationships with public companies.
The state also invests in AI companies directly. The central government operates several technology investment vehicles. Local governments are increasingly creating their own, often armed with billions of dollars. Tianjin, a coastal metropolis, announced a $16 billion investment AI funds in 2018.
Public capital is increasingly helping to fill the void left by foreign investors spooked by US sanctions against some Chinese companies. AI darlings, considered too close to the Communist Party. A fund run by the Cyberspace Administration of China, a regulator, has acquired an undisclosed stake in SenseTime, which was hit with a new round of US sanctions last month for its alleged involvement in a government crackdown on the Uyghur ethnic minority. (SenseTime says the penalties are based on a “misperception” of its business.) A separate vehicle, the Mixed Ownership Reform Fund, accounted for $200 million of the $765 million the company raised in of its IPO (Initial Public Offering). Local governments contributed an additional $220 million.
lost in translation
State dosh, combined with access to lots of public data, helped transform the Chinese AI powerful companies in certain niches. According to Bain, a consulting firm, last June the cloud division of Alibaba, the Chinese e-commerce giant, offered 62 AI-enabled services, from voice recognition to video analysis, compared to 47 for its closest Western rival, Microsoft. SenseTime and Megvii mass-produce computer vision software and hardware that can be adapted and installed in individual factories. Despite being barred from most Western markets by US sanctions, SenseTime garnered 762 million yuan in overseas revenue in 2020, up from 319 million yuan two years earlier, mostly in Southeast Asia. East.
Despite all these successes, China AI the industry is following the West in a big way. Although it leads America in the total number of AI-, China produces fewer peer-reviewed papers that have academic and corporate co-authors or are presented at conferences, which are usually held at a higher level. It ranks below India, and well below America, in number of skilled workers AI coders relative to its population. These shortcomings are likely to persist, for three reasons.
First, capital may not be allocated efficiently. It’s unclear, for example, how much of Tianjin’s $16 billion prize pool has actually been deployed. More damagingly, Beijing has created a reward system for local officials that favors debt-fueled spending and rarely punishes waste.
Lots of state AI the investments have been “irresponsible and redundant”, says Jeffrey Ding of Stanford University. Lancaster University’s Zeng Jinghan has documented the rise of companies that falsely claim to grow AI in order to aspire subsidies. An analysis by Deloitte, a consulting firm, estimated that 99% of self-declared people AI startups in 2018 were fake. Such waste not only burns public money, Ding notes, but also consumes scarce human capital that could have been more usefully deployed elsewhere.
China’s second problem is its inability to recruit the best in the world AI minds, especially those working on high-level research. A 2020 study by MacroPolo, a Chicago-based think tank, showed that more than half of top researchers in the field worked outside of their home country. America and Europe seem more attractive to these mobile brains, including many Chinese. Although about a third of the best in the world AI talent comes from China, only a tenth actually works there. A shortage of non-Chinese scholars further handicaps China’s capabilities, notes Matt Sheehan of the Carnegie Endowment for International Peace, a think tank in Washington.
Even more problematic for the party, its master plan ignored the advanced semiconductors that power AI. Since its release, Chinese companies have been finding it increasingly difficult to get their hands on advanced computer chips. Indeed, virtually all of these microprocessors are either American or made with American equipment. As such, they are subject to the restrictions on exports to China put in place by Donald Trump and extended by his successor to the presidency, Joe Biden. It will take years for Chinese companies to catch up with the global edge, if they can.
These challenges will continue to plague all of China’s high-tech industries for years to come. He could leave his AI companies stuck in a rut – successfully deploying relatively unsophisticated products while following Europe and America in breakthrough developments of greater financial and strategic value. Consider Wu Dao 2.0. Although this was a huge improvement over Google Tags-3 is exactly what he did: improve an existing technology rather than innovate. No amount of Chinese taxpayers’ money is likely to change that. ■
This article appeared in the Business section of the print edition under the headline “In pursuit of mastery”