Receive free technology regulation updates
I will send myFT Daily Digest E-mail summarizing the latest information technical regulation News every morning.
The Chinese government plans to issue rules on generative AI, aiming to strike a balance between encouraging local companies to develop innovative technologies and China’s desire to control content.
The China Cyberspace Administration, a powerful internet watchdog, aims to create a system to force companies to license generative AI systems before releasing them, two people close to the regulator said.
The requirement strengthens a draft regulation issued in April that gave the group 10 working days to register its products with Chinese authorities after launch, giving the group room to act. .
The new licensing scheme is part of a set of regulations that will be finalized as soon as this month, according to people familiar with the move, and how the Chinese government will reconcile its longstanding censorship regime with its ambition to develop world-leading technology. It shows how hard you are trying to get.
“This is the first time [authorities in China] Matt Sheehan, a researcher at the Carnegie Endowment for International Peace, said there was a “trade-off” between two Communist Party’s “fundamental” goals of maintaining AI leadership and controlling information.
“If the Chinese government intends to fully control and censor information generated by AI, it will require all companies to seek prior approval from the authorities,” said a person close to the CAC deliberations.
“However, regulations must avoid stifling domestic firms in the technology race,” the official added. The authorities are “shaking”.
China is seeking to respond to the rise of generative AI systems that rapidly create human-like text, images, and other content in response to simple prompts.
According to the draft rules from April, content “embodies core socialist values” and “could overthrow state power, advocate the overthrow of the socialist regime, or divide the country.” It should not contain any content that “incites the national unity or undermines national unity.”
Companies such as Baidu and Alibaba that have deployed generative AI applications this year have been in contact with regulators over the past few months to ensure their AI does not violate rules, according to two people close to the regulator. It is said that he was taking
China’s government has concerns about the data being used, so the CAC needs to ensure AI is “reliable and controllable,” its director Zhuang Rongwen said recently.
“China’s regulatory measures are mainly focused on content control,” said Angela Chan, associate professor of law at the University of Hong Kong.
Other governments and authorities are also rushing to enact legislation against potential abuses of technology. The EU has proposed some of the world’s toughest rules, drawing backlash from the region’s businesses and executives, while the US government is debating AI control measures and the UK is undertaking a review.
China’s April regulatory draft sets out requirements for the data tech companies use to train generative AI models based on specific demands to ensure “accuracy, precision, objectivity and diversity.” rice field.
The requirement shows that China is adopting a similar direction as Europe, where AI systems are used to train AI models, such as tackling problems such as “hallucinations” in manufacturing materials. The quality of data collected has become an important area of regulatory scrutiny.
But the Chinese government has set the requirements “very high,” Sheehan said. This means that Chinese companies will have to put more effort into filtering the types of data used to “train” their AI.
However, the lack of available data to meet these demands has prompted many to develop and improve so-called large-scale language models, the technology underlying chatbots such as OpenAI’s ChatGPT and Google’s Bard. It is a bottleneck that prevents companies from doing so.
Businesses ‘likely to be more cautious and conservative about what’ [AI] Because the consequences of breaking the rules can be severe,” said Helen Toner, director of strategic and basic research grants for Georgetown University’s Center for Security and Emerging Technologies.
Chinese officials have implied in the draft regulation that the technical groups creating AI models will be held almost entirely responsible for any content created, which means that “companies cannot control They may be reluctant to publish their models because they may be held accountable for issues they don’t have.” ‘ said Toner.
The FT has reached out to the CAC for comment.
Additional report from Ryan McMorrow in Beijing
