Algorithms trained on mountains of Chinese data may soon be making decisions that deeply affect the lives of people in the US.
Take Yitu Technology, a Shanghai-based artificial intelligence start-up that won top honours in two AI competitions in the US last year for its facial recognition technology . The system was built for Chinese law enforcement using data collected by the authorities. Or as the company boasts, it was honed on the “world’s largest portrait system, covering more than 1.5bn people.”
Yitu is now looking for customers in the US to put its software to work. “There are a lot of applications for this technology,” says Wu Shuang, who heads its Silicon Valley research group.
This story is the first in a series on the growing competition between the US and China over AI, which will have an impact on the contest for military superiority, the international trading system and the future of the Chinese Communist party.
It is not alone. Shenzhen-based Malong Technologies has also trained its image recognition algorithms on masses of Chinese data — in its case, by analysing hundreds of thousands of photos from fashion shows to identify trends for clients in the garment industry. It says it is now trialing the technology with ecommerce companies in the US.
A “key difference in China is there are just more people, more data, more businesses — it’s just bigger,” says chief technology officer Matt Scott, a former Microsoft researcher who moved to China to co-found the company. “Having access to that data in China, we can export [the technology] around the world.”
Algorithms like these are the advanced guard in a battle that will go a long way to determining economic leadership in the era of Big Data — a contest where China is catching up quickly and now vying with the US to be the dominant force.
The AI revolution is often thought of in terms of robots or drones that can do tasks once performed by humans. But its impact will also be felt from a less visible source — the ability to sweat the data the hardest. Machine learning systems that can find patterns by analysing large data sets are at the cutting edge of today’s artificial intelligence.
For some industries, deep learning — the most advanced form of the technology — has the potential to create value equivalent to as much as 9 per cent of a company’s revenues, according to a report in April from McKinsey Global Institute. That translates into trillions of dollars of potential economic value — and the US and China are the clear leaders.
“If you look globally, it’s a two-horse race in AI,” says Michael Chui, a McKinsey partner who led the study.
In China, the AI boom has fed the country’s swelling sense of self-confidence in its expanding technology base. President Xi Jinping has made AI one of the central pillars of the Made in China 2025 plan to transform the country’s economy and has set a goal of being the world leader in the technology by 2030.
At the same time, China’s advances are also contributing to an opposite paranoia in the US that its technology exceptionalism can no longer be taken for granted. The Trump administration’s plans for a trade war with Beijing are motivated — at least in part — by fear of China’s advances in new technology.
“It’s clear that the US government sees itself in a tech arms race with the Chinese government,” says Robert Silvers, a partner at legal firm Paul Hastings and former assistant secretary for cyber policy at the department of homeland security. “The US is taking the view that these kinds of technologies are so transformative that the country that gets the lead is going to have not just economic or tech advantage but also national security advantage.”
One reason the contest over AI is so charged is that it is connected with a race to find a new military edge. As well as answering mundane customer queries and piloting driverless cars, the same technology can also be deployed to synchronise drone swarms, analyse pictures taken by spy drones and control autonomous boats.
Dominance in AI could bring a step change in warfare, says Sean Gourley, founder of Primer, a Silicon Valley AI start-up whose backers include the CIA’s venture capital arm. Technology shifts like this can undermine the military advantage of great powers.
“It’s likely to be coupled with the reordering of global power. Whoever is best at this will be in a strong position in 10 years’ time,” he says.
Russian president Vladimir Putin found his own way to raise the rhetorical stakes over AI last year: “Whoever becomes the leader in this sphere will become the ruler of the world.”
According to most experts, the US still has a clear lead. It takes three things to be a world-class AI power: the most advanced algorithms, specialised computing hardware, and a good supply of the raw material that machine learning systems depend on — data.
Last year’s Go match, where a system built by Google subsidiary DeepMind trounced leading player Ke Jie in the ancient Chinese board game, was China’s wake-up moment in AI, says one Google executive. “Only when the Russians launched the sputnik did the US realise how far they had come,” he says. “China had that moment when they lost at AlphaGo.”
That view is echoed by some in China. “For top talents, clearly the US will still be the main resource. I think there’s no question of that,” says Rong Jin, head of machine intelligence technologies at DAMO Academy, Alibaba’s research programme. The perception in China is that Americans throw themselves into fundamental research and are heavy duty mathematicians — the disciplines at the heart of AI — while Chinese tend to study coding or engineering.
Yet despite those advantages, China is rapidly narrowing the algorithm gap. When it comes to the output of China’s research institutions, “the statistics are definitely rising sharply”, says Oren Etzioni, who runs the AI research institute of Microsoft co-founder Paul Allen. He points to other signs of China’s growing AI capability, from the reading comprehension test earlier this year in which AI newcomer Alibaba tied for top honours with traditional research power Microsoft, to the strong showing of Chinese researchers in the annual ImageNet competition for image recognition.
On the second category of hardware development, China has been slower to build the sort of homegrown chip industry needed to put it on the leading edge. That has been partly due to a series of decisions that effectively bar the acquisition of US chip companies, which started under the Obama administration and accelerated under President Donald Trump.
It is in the final area — the availability of raw data — where most experts believe China’s AI advantage lies.
China has reams of data on its citizens and is not afraid to use it. This is partly due to a state that monitors everything from birth: facial recognition is so widespread you can be picked up for jaywalking and stopped from stealing tissue at the Temple of Heaven in Beijing.
But it is also a tribute to China’s early move online: this is a country where people order, shop, pay and play online, leaving massive data footprints that enable merchants to accurately target ads and promotions. “The density of people is proportional to the density of data,” says a leading Chinese machine intelligence scientist.
Chinese attitudes to data privacy are becoming slightly less lax, but regulations are still a million miles from Europe, which is at the other end of the spectrum and will introduce tough privacy rules later this month known as General Data Protection Regulation. Yet American companies like Facebook, Google and Amazon also have masses of data, says Mr Wu at Yitu.
That suggests that general-purpose AI applications like facial recognition will be the preserve of all “the big platforms”, regardless of their country of origin, says James Manyika, a partner at McKinsey. By contrast, more specialised applications could be perfected where the data are the richest. When it comes to manufacturing, for instance, China is “collecting a lot more data”, he says.
This data advantage could be greatest in fields where regulation has made access to information harder, or prevented it being collected in the first place, according to some experts. Earlier this year, Google published promising research suggesting it could predict the risk of heart attack by using image-recognition software to study retinal blood vessels. The research relied heavily on UK Biobank, a database drawing on a detailed study of volunteers in Britain beginning in 2006.
Yet only 631 people in the Biobank had medical conditions relevant to the research. That made the data set “relatively small for deep learning”, Google said, reducing the effectiveness of the algorithm it was able to train on the information. Chinese medical AI researchers, by contrast, have been able to tap into far bigger data sets, according to one expert.
If China is rich in data, then it also has the economic opportunities to exploit it — something that has helped lure back many haigui, or returning “sea turtles”, like Mr Jin. AI is being harnessed in law, where machines have replaced stenographers in more than 6,000 courts; on the roads to manage traffic; in hospitals to detect tumours; and in Shanghai subway stations where you can buy tickets by talking to the machine.
“AI has the biggest opportunity in China versus any western countries,” says Mr Rong.
Chinese executives talk about a smart city scheme that halved the time it takes to speed ambulances from depot to patient to hospital, by dint of switching traffic flows and traffic lights.
The smart cities scheme also offers another point of difference with the US: collaboration between state and private companies on a large scale. In addition to projects aimed at traffic management, crowd control, finding missing children and elderly, cutting down hospital waiting times — the list goes on — all the big tech players have joint research labs with government.
This is part of a broader experimentation that is lacking in the US, says Mr Wu. “Overall, the Chinese tech scene is more dynamic right now, particularly in terms of trying out new ideas and new products,” he says. “People are just trying out more new things.”
That has not been lost on investors in the US. One leading Silicon Valley venture capitalist puts the difference in AI opportunities bluntly: “The business is bigger and better in China.”
This economic momentum behind AI is closely aligned with a second powerful force: a sense of national mission. That has brought a hydrant of money and clear industrial policy. This state-led strategy is also closely aligned with national champions Baidu, Alibaba and Tencent — all private companies.
Washington has done much less to promote a national agenda. “What’s the national direction around AI and robotics [in the US]? It’s nothing. It’s missing,” says the Silicon Valley investor. “The government is flailing around.”
Worse, the Trump administration’s attempts to clamp down on immigration has upset the US tech industry, which has drawn heavily on overseas talent — not least Indian and Chinese engineers. The heads of AI at Apple, Facebook and Microsoft, as well as Google’s cloud computing division, were all born outside the US.
“We’ve seen more and more students choosing not to come to the US,” says Mr Etzioni. “We’re in the process of shooting ourselves in the head.” He points to one sign of how the talent pendulum is swinging away from the US: Google and Microsoft have both opened AI research centres in China to tap the AI workforce there.
Yet the expertise advantage that the US has will not disappear overnight. Companies like Yitu are moving in the opposite direction because they believe the US west coast is still the magnet for many of the world’s top engineering brains. “Half the AI engineers in Silicon Valley are Chinese,” says Mr Wu.
This article was updated on May 1 to correct Matt Scott’s role at Malong
Young sector uses open-source technology
For some, artificial intelligence is a new arena for geopolitical competition with military overtones. But for others, it is the first truly open-source technology, with companies and nations alike sharing ideas to improve the lot of humankind.
“It’s probably the best worldwide collaboration anyone has ever seen,” says Kai-Fu Lee, who headed up Google’s China operations and now runs his own venture capital firm.
Many tech companies like to make the same argument. “We hope we can develop AI tech and share with all players,” Tencent chief operating officer Ren Yuxin told a gathering of developers last year in Chengdu. Tencent and Google last year agreed to cross-license patents on a range of products and technologies.
“AI is mostly an enabler, it’s not a nuclear weapon,” says Mr Lee. “I worry about the cold war analogy.” He says it is being firmly driven by universities and tech giants rather than defence departments and any competition is at company rather than national level.
James Lewis at the Center for Strategic and International Studies in Washington agrees that the competition “is largely commercial and it’s largely between companies.”
Proponents of this view point to the open-source resources and willingness of scientists to upload their findings immediately rather publish in journals.
Data are not always a company-specific resource, but can be bought from providers — a classic gig economy business that uses shoals of workers to provide and label speech, image and other data in a way a computer can understand.
“There are reams and reams of data,” says Mark Brayan, chief executive of Appen, a Sydney-based company. “But is it the right data? Is it prepared and labelled in the right way? Is it of sufficient quality?”
Copyright The Financial Times Limited 2018
2018 The Financial Times Ltd. All rights reserved. Please do not copy and paste FT articles and redistribute by email or post to the web.
