China is set to require a security review of generative artificial intelligence (AI) services, like ChatGPT, before they can operate. The move aims to protect the intellectual property of the data used by AIs and prevent discriminatory bias.
The Cyberspace Administration of China (CAC) has released draft guidelines for public consultation, which state that service providers must ensure their content is accurate and respects intellectual property. They must also ensure it does not discriminate or jeopardize security. Additionally, AI operators must clearly classify AI-generated content, according to the country’s internet watchdog.
These requirements are part of China’s attempts to regulate the explosive growth of generative AI since ChatGPT, created by OpenAI, shook the industry in November. Companies like Alibaba, SenseTime, and Baidu aim to build the ultimate next-generation AI platform for the world’s largest internet marketplace. This initiative reflects a wave of projects abroad, with Google, owned by Alphabet, and Microsoft among the many tech companies exploring generative AI.
It remains unclear how the government intends to promote and police the new course, although the guidelines released on Tuesday essentially carry over existing regulations on data and content into the burgeoning field of AI.
It is likely that China will bar foreign AI services, such as those from OpenAI or Google, as it has done to US search engines and social networks. The Beijing government insists on strict control over online content and discussion. However, for now, China is expected to avoid tightening its grip on local companies too much, for fear of crushing a nascent arena that needs room for innovation.
On Tuesday, e-commerce giant Alibaba outlined how it plans to build generative AI into its Slack-like enterprise messaging app and Amazon Echo-style smart speakers, before expanding that portfolio to its other services. A day earlier, SenseTime had introduced the great SenseNova AI model and a user-facing chatbot called SenseChat. These projects followed Baidu’s Ernie bot, which was released for selective testing about a month ago.
The new regulations are likely to interfere with how AI models in China are trained going forward. “There is real potential to affect how models are trained, and that stands out to me as very important here,” said Tom Nunlist, senior analyst at Trivium China. However, much depends on how regulators interpret the general provisions in their notice.
The Beijing government plans to introduce rules to oversee the use of artificial intelligence in various industries. In February, officials from the Science and Technology Ministry said the government would push for the safe and controllable application of AI services, which it considers a strategic industry.
Alibaba shares returned much of their early gains on Tuesday after the CAC announcement, while SenseTime shares barely moved. Stocks from Chinese carriers such as China Mobile and smaller AI companies such as 360 Security Technology and AI chipmaker Cambricon Technologies also showed sharp declines.
