It’s definitely impressive — but it’s not quite “Big Brother” yet. REUTERS/Bobby Yip
The Chinese government is working to create a techno-authoritarian state powered by artificial intelligence and facial recognition to track and monitor its 1.4 billion citizens.
The government has big plans to have a ubiquitous surveillance network, leading the country to becoming the biggest market in the world for video surveillance — $6.4 billion in 2016, according to estimates from IHS Markit Ltd. China already has 170 million security cameras in use for its so-called Skynet surveillance system, with 400 million more on the way in the coming years.
Far from hiding its wide-reaching abilities from the public, the government has frequently touted its high-tech surveillance successes in recent months.
Last September, English-language state newspaper China Daily touted how police in Qingdao used facial recognition technology to catch 25 would-be criminals. In March, Beijing police began using facial recognition and AI-powered glasses to catch criminals — just a couple months after police in Henan and Zhengzhou began testing the glasses at train stations.
In Xiangyang, a giant screen was set up over a crosswalk to display the names and faces of jaywalkers and other lawbreakers that cameras caught at the intersection. And in December, China demonstrated its sophisticated “Skynet” system by having it track down a BBC reporter in just 7 minutes.
But all of these successes belie a simple reality: the surveillance tech is not nearly as pervasive or effective as the government or the media purports it to be.
Face++ isn’t all-powerful yet
Entrance to Megvii’s offices are managed by its Face++ software.Harrison Jacobs/Business Insider
On a recent visit to the offices of Megvii, a leading artificial intelligence startup and one of the main providers behind the facial recognition tech used by Chinese police, I met with Xie Yinan, the company’s vice president.
Despite notions that Chinese police’s facial recognition capabilities can track down anyone, anywhere, that’s simply not what the technology is capable of, according to Xie.
He said Megvii’s Face++ platform, which numerous police departments in China have used to help them arrest 4,000 people since 2016, has serious technological limitations.
For example, even if China had facial scans of every one of its citizens uploaded to its system, it would be impossible to identify everyone passing in front of a Face++-linked camera. While the Face++ algorithm is more than 97% accurate, it can only search a limited number of faces at a time.
In order to work, police would have to upload the faces they want to track to a local server at the train station or command center where they intend to look. Face++ would then use its algorithm to match those faces to the ones it encounters in the real world.
Harrison Jacobs/Business Insider
Xie said it wouldn’t be feasible to have the system search for more than 1,000 faces at a time — the data and processing power required for an operation larger than that would require a supercomputer. Plus, Xie said they can’t run the system 24/7 today. It’s the kind of thing police will have to activate proactively when a situation is underway.
While it is possible that the system could be connected to a supercomputer over the cloud to amplify computing power, it would be too dangerous from a security perspective. The system has to stay offline and local.
When I asked whether Xie or the company have any concerns over how police could misuse the Face++ platform, he essentially said it’s up to the government to write the legal framework on when and how law enforcement can use it.
“We don’t have access to the data,” he said. “What we do is sell them a server [loaded with Face++]. That’s all.”
Exaggerating technological advancements
Facial recognition isn’t the only area where China’s techno-authoritarian capabilities have been exaggerated, by both the media and the government.
At the crosswalk in Xiangyang, there is a 5- to 6-day delay between when someone commits crime and when their face appears on the billboard. Local officials told The New York Times that humans, not an algorithm, look through the photos the crosswalk camera captures to match them with people’s identities.
Meanwhile, the smart glasses police are using in Beijing and Zhengzhou only work if a target stands still for several seconds. It’s less being used to spot criminals than to verify travelers’ identifications.
Police in China monitor vast quantities of surveillance cameras from central command hubs. AFP/AFP/Getty Images
But, in some ways, it hardly matters. Those nuances are often lost on the public, particularly when state media has gone to such lengths to convince its populace of its technological prowess.
In Zhengzhou, a heroin smuggler confessed after police showed the suspect their smart glasses and said it could incriminate him, The Times recently reported.
“The whole point is that people don’t know if they’re being monitored, and that uncertainty makes people more obedient,” Martin Chorzempa, a fellow at the Peterson Institute for International Economics, told The Times.
Of course, it’s likely only a matter of time before the technology gets better. The Chinese government and the country’s tech investors are pouring money into facial recognition startups like Megvii.
Megvii raised $460 million last November, much of which came from a state-owned venture fund. While the valuation hasn’t been disclosed, it’s likely that it is close to or tops $2 billion. Two smaller Chinese companies include DeepGlint, and Yitu Technology, which raised $380 million last year.
SenseTime, a competitor, became the world’s highest valued AI startup after raising $600 million in April and $620 million in June. It now has a valuation of $4.5 billion.