China announces rules to keep AI bound by ‘core socialist values’
The “Interim Measures for the Management of Generative Artificial Intelligence Services” announced Thursday and set to take effect on Aug. 15 represent Beijing’s attempt to encourage the growth of China’s AI industry while retaining total control over information available to the public. It is an enormous challenge made more difficult by the rising global popularity of tools that allow people to generate unique text, images and music.
For now, Beijing has landed on an answer: companies and research labs can innovate, but strict rules apply to public-facing services.
Analysts criticized an earlier draft of the regulations released in April as deeply unfriendly to the industry. Some requirements, they said, such as that companies should verify the accuracy of the data their AI models learned from — which in many cases includes huge chunks of text from the internet like Reddit and Wikipedia, both banned in China — would be nearly impossible to comply with.
The final rules have walked back some of these controls in specific settings, like AI research labs and companies that make AI tools used by other businesses.
But companies providing AI-generated content to the public must still take steps to ensure such content is accurate and reliable — and in line with China’s values.
“The measures clearly state that the provision and use of generative artificial intelligence services should adhere to the core socialist values,” the internet regulator Cyberspace Administration of China said in a statement.
Experts say this places a heavy burden of compliance on the dozens of companies that are rushing to develop tools like ChatGPT for the Chinese public.
The measures add to a network of existing regulations on China’s tech companies. Under Beijing’s censorship regime, tech companies such as Tencent, ByteDance and Weibo bear much of the responsibility for monitoring their own platforms.
The new rules will make companies that offer generative AI services to the public similarly responsible if anything their products create causes trouble, said Helen Toner, director of strategy at Georgetown University’s Center for Security and Emerging Technology.
“This is a pretty significant set of responsibilities, and will make it hard for smaller companies without an existing compliance and censorship apparatus to offer services,” Toner said.
In chatting with Ernie, Chinese search giant Baidu’s prototype AI chatbot, The Washington Post found that even simple requests for facts about China’s government or top leader Xi prompted it to terminate the exchange with a canned reply.
One effect of these challenges could be that Chinese tech companies decide to focus on developing AI products for other companies to use, rather than for the public.
The rules come as Beijing has signaled an end to a crackdown on the technology sector that has wiped more than a trillion dollars in market capitalization from China’s biggest tech companies since 2020, by some estimates.
The crackdown, which kicked off with the shelving of mobile payment titan Ant Group’s planned $37 billion IPO, reined in the country’s tech giants, including investigating ride hailer DiDi Chuxing and banning the lucrative online tutoring industry. It culminated with a $985 million fine against Ant Group earlier this month.
The rules on generative AI explicitly state Beijing’s support for the sector and interest in Chinese companies’ ability to compete with their foreign counterparts on the world stage.
Zhou Hongyi, billionaire founder of security company Qihoo 360, said in a post on China’s Twitter-like site Weibo that regulators had been “very willing to listen” to tech companies’ input on the rules.
The measures “provide the confidence and assurance that technology companies need to innovate,” said Zhou.
The rules also represent one of the most detailed efforts yet by any government to regulate AI as lawmakers around the world grapple with the challenge of guarding against potential harms, including protecting personal information and intellectual property.
But with the word “interim” in the title, further regulation of the sector is sure to follow. One provision also calls for participation in forming international rules on generative AI, something that Elon Musk had said Chinese officials expressed interest in during his recent visit to Beijing.
“China is willing to enhance communication and exchanges with the international community on AI security governance, promote the establishment of an international mechanism with universal participation, and form a governance framework and standards that share broad consensus,” said Foreign Ministry spokesman Wang Wenbin on Thursday.
Pei-Lin Wu in Taipei contributed to this report.
