China faces a problem familiar to dictatorships throughout history: how to strike a balance between the growth-friendly innovation that thrives in a free society and the paranoia of an authoritarian state. China's top leader, Xi Jinping, wants the country to become a hyper-advanced economy. His government actively promotes the commercialization of the advanced technologies it likes, from electric vehicles to quantum computing.
At the same time, he dedicates himself to tightening the screws on those he doesn't like. In 2021, he sank a burgeoning online education sector almost overnight with regulations; apparently for fear that its high costs would make raising children so expensive that the Chinese would give up on the idea of being parents. On December 22, the government dealt another blow to the video game industry by introducing rules to, among other things, limit player spending on in-game purchases and, therefore, developers' profits. The market value of Tencent, one of China's most innovative companies that also has a large gaming business, plummeted 12%.
Nowhere is this tension more clearly reflected than in the most disruptive technology of 2023: artificial intelligence. In many countries, control of artificial intelligence is considered economically and strategically important. Around the world, politicians fear that machines will become hostile or, more realistically, used by malicious human agents. In Beijing, the added worry is that the technology – whose flourishing depends on unlimited data and, in its current phase of development, unregulated spaces – will end up being subversive if not kept under control. For this reason, the State is working hard to prepare its Great Firewall for the age of artificial intelligence.
In 2000, Bill Clinton, then president of the United States, compared China's efforts to control the Internet to "trying to nail a gelatin pudding to the wall." Today the gelatin seems to remain firmly in place. Western Internet services, from Facebook and Google search to Netflix, are not available to most Chinese (except for those who agree to take the risk of using illegal "virtual private networks"). On local platforms, all undesirable content is eliminated, either preventively by the platforms themselves using algorithms and armies of moderators, or a posteriori as soon as it is detected by state censors. In 2020, a series of severe measures brought the powerful Chinese technology giants (such as Tencent and Alibaba) into line and brought them closer to the State, which has been acquiring small stakes in the companies and showing great interest in their daily operations.
The result is a sterilized and prosperous digital economy. WeChat, Tencent's super app that combines messaging, social media, e-commerce and payments, alone generates hundreds of billions of dollars in annual transactions. Xi now hopes to strike a similar balance with artificial intelligence. Again, some foreign experts predict a “jelly” situation. And, again, the Communist Party is creating tools to prove them wrong.
The party's efforts have begun with establishing the world's strictest standards for Chinese equivalents of ChatGPT (which, predictably, is banned in China) and other consumer-oriented "generative" artificial intelligence. Since March, companies have had to register with the authorities all algorithms that make recommendations or that can influence people's decisions. In practice, that basically means all such software aimed at consumers. In July, the government enacted rules requiring all AI-generated content to "uphold socialist values"—that is, not include obscene songs, anti-party slogans, or, God forbid, mockery of Xi. In September, published a list of 110 registered services. Only its developers and the government know all the details of the registration process and the precise criteria followed.
In October, a national information security standards committee published a list of security guidelines that require detailed self-assessment of the data used to train generative AI models. One standard requires manual testing of at least 4,000 subsets of the total training data; At least 96% of them must be rated "acceptable," based on a list of 31 loosely defined security risks.
The first criterion for unacceptable content is anything that "incites the subversion of state power or the overthrow of the socialist system." Chatbots must refuse to respond to 95% of queries that raise unacceptable content (and presumably those that do manage to pass that filter will be subsequently censored). At the same time, they should reject no more than 5% of harmless questions.
Everything produced by unregistered algorithms will be blocked, and its creators will be sanctioned. In May, a man from Gansu province was arrested after using ChatGPT to produce text and images of a fake train accident and posting it all on social media. He may have been the first Chinese arrested for spreading disinformation generated by artificial intelligence. He won't be the last.
Such a heavy-handed strategy has slowed the implementation of consumer-oriented generative artificial intelligence in the country. Ernie Bot, created by Baidu, another tech company, was ready around the time ChatGPT launched, but it didn't go on sale until nine months later in August – an eternity considering how quickly it evolves. the technology. The chat continues to be clumsy when it comes to expressing its devotion to the game. When faced with sensitive questions related to Xi, he obediently censors, offers no response and deletes the query.
With a little more work, the party may not only be able to turn artificial intelligence models into good communists, but also into communists capable of expressing themselves with ease. That would avoid the need for post-facto censorship, says Luciano Floridi of Yale University. In any case, the authorities do not seem to be in a hurry to get to that point. Instead, they are dedicated to promoting business applications of the technology.
Unlike consumer AI, the development of enterprise AI faces few limitations, says Mimi Zou of the Oxford Martin School research institute. The consequence, as Steven Rolf of the Center for Research on Digital Futures at Work, a British think tank, explains, is that capital and labor are moving away from things like consumer chatbots and towards machine learning to the companies. This, the government seems to consider, will allow China to catch up and even surpass the United States in artificial intelligence, without the inconvenience of having to face the potentially subversive content generated by it.
In May, the southern city of Shenzhen announced it would launch a 100 billion yuan (€13 billion) artificial intelligence investment fund, the largest of its kind in the world. Several municipal governments have launched similar funds. Much of that money goes to companies like Qi An Xin, which offers generative artificial intelligence that manages risks related to enterprise data security. According to the company, the bot can do the work of 60 security experts, 24 hours a day. Before going public in 2020, it received large investments from state-owned companies, like many other similar startups.
For that strategy to work, enterprise AI companies need the right raw materials. Consumer chatbots use artificial intelligence models trained with large swathes of the public Internet. Corporate applications need corporate data, much of which remains hidden within companies. So the other pillar of China's strategy is to turn corporate data into a public good. The State does not want to own the data, but rather (as in the case of the other factors) control the channels through which it flows.
To do this, the government promotes data exchanges. The aim is for different businesses to be able to exchange information packaged in standardized products about all areas of commercial life, from activity in specific factories to sales data in specific establishments. Small companies would thus gain access to knowledge previously reserved for technology giants. Banks and intermediaries would have a real-time image of the economy.
Chinese cities began launching data exchanges about a decade ago. There are currently about 50 throughout China. And they are finally picking up speed. The Shanghai Data Exchange (BDS), which was launched in 2021, has started trading new data products. In one of its first transactions, ICBC bank purchased information from the energy sector. Such information can be used for the purposes of assessing companies' energy consumption and, to the extent it reflects actual activity levels, creating alternative credit profiles for companies. The BDS sells satellite data on steel production in central China and environmental violations by mining companies. Another product provides real-time data on doctors, nurses and hospital beds across the country to help medical companies make business decisions. The BDS is also experimenting with using data as collateral to obtain loans.
On a large scale, the datafication of the industry could generate a significant economic boost, says Tom Nunlist of Trivium, a Beijing consultancy. And it may not be long before the exchanges receive more data. In August, the central government tasked state-owned companies with studying how to value their data. In recent months, teams of auditors have tried to find a way to add them as intangible assets to companies' balance sheets. They are supposed to submit the reports by January 1 (although the deadline is likely to be missed, given the unprecedented nature of the task).
Now, the government's commitment to business artificial intelligence is not without problems. The automobile industry is a good example. In 2022, some 185 million vehicles with an Internet connection will circulate on Chinese roads, and there is a national plan that foresees the mass production of semi-autonomous cars by 2025. For that to happen, companies dedicated to the design of autonomous driving algorithms need lots of data with which to train your systems. A company called WICV is building a platform for the data that's starting to come out of cars.
For now, WICV returns a car's data to its manufacturer, so BYD gets data from BYD cars, Nio gets data from Nio cars, and so on; but the plan is that, over time, the data will be traded on exchanges, where it can be purchased by other developers of autonomous driving systems. However, driving data will have to be "desensitized" first, explains Chu Wenfu, founder of WICV; That is, removing biometric and geolocation details that could help malicious actors track the movements of specific people.
The potential for such tracking scares Chinese authorities. In 2021, one of the main reasons for the crackdown against Didi Global days after the ride-hailing company's listing on the New York Stock Exchange was, it later emerged, fear that data on the 25 millions of daily Didi trips, including geolocation information linked to passengers, could fall into the hands of US authorities. The Chinese government is preemptively identifying vast areas of China where data collection could pose a threat to national security.
Many automakers, including Western ones like BMW, have no choice but to partner with state-backed companies to manage driving information and ensure local data compliance rules are met. Just in case, some car companies are dropping some features, such as allowing drivers to view live images of the interior and exterior of their car on their phones. After all, some of those recordings could inadvertently capture something sensitive.
Such trade-offs between innovation and safety are unlikely to be limited to cars. Other industrial and corporate data will likely also need to be desensitized before it can be traded on a large scale on exchanges. That will slow the development of enterprise artificial intelligence, although the algorithms remain unconstrained. It is a price the party seems willing to pay for its paranoia.
© 2023 The Economist Newspaper Limited. All rights reserved
Translation: Juan Gabriel López Guix
