The Chinese government just released a new “Global AI Governance Initiative,” which specifically states that China “[opposes] using AI technologies for the purposes of manipulating public opinion, spreading disinformation, intervening in other countries’ internal affairs, social systems, and social order, as well as jeopardizing the sovereignty of other states.” For anyone worried about the ever-growing amount of disinformation online, and specifically the constant stream of public research that points to China’s embrace of just such tactics, China’s public pledge should come as a relief.
There’s just one problem: China is lying.
In fact, Chinese Party–state actors are already embracing using generative AI to create and spread disinformation, and have long been using earlier forms of AI for this same purpose. In September, Microsoft reported that China-affiliated actors began experimenting with text-to-image models in, at the latest, March 2023, and that one of the images produced and circulated was a distorted image of the Statue of Liberty. The Microsoft report came after reports throughout the last several years that China has been using AI-generated avatars to spread propaganda.
Such reports confirm many of the worst-case fears among those of us—myself included—who study China and disinformation, and have been arguing for awhile now that generative AI is a dream come true for Chinese propagandists, and that this technology will likely be quickly adopted, and push into hyperdrive China’s efforts to shape the global conversation. A recent State Department report (PDF) on China’s efforts to reshape the global information environment further bolsters our concerns.
China’s plan to use AI for disinformation is obvious, if you know where to look. Li Bicheng, a Chinese military-affiliated researcher, has been designing a system for “online public opinion guidance” and employing “online information deception” since at least 2016. In 2019, he proposed using AI to make his disinformation system more authentically human-seeming in several ways, such as by creating a network of “intelligent agents,” or AI-run, fake accounts that could push pro-CCP messages on social media. As recently as January this year, Li was working on research, directly funded by the Chinese military, to create more “authentic” text with AI, text that will better reflect natural human language.
Few in the Biden administration will be surprised to learn that Beijing is lying about its intentions and activities when it comes to influence operations. But lies, maybe even especially blatant ones, present diplomatic opportunity. The U.S. government should call Beijing’s bluff, and propose to discuss the issue when President Biden and Chinese General Secretary Xi Jinping meet Wednesday, in San Francisco, during the Asia-Pacific Economic Cooperation summit.
Ideally, the two leaders would agree on a bilateral pledge along the line of China’s proposal, forswearing using AI to manipulate or otherwise influence the other country’s population. China’s track record of abiding by these types of pledges is very poor, evident in the failure of similar agreements for China to stop the cybertheft of intellectual property from U.S. companies under the Obama administration. However, the purpose of this pledge would be to hold Beijing accountable publicly in the future, and to make clear to Xi in person that the United States opposes such influence operations.
Such a pledge would come at an especially critical moment, as the U.S. government is worried about the potential for Chinese election interference in the upcoming 2024 elections. The U.S. intelligence community warned earlier this year (PDF) that China has “shown a willingness to meddle in select election races that involved perceived anti-China politicians,” and other U.S. government officials have made similar warnings about potential Chinese interference in upcoming elections.
The rise of generative AI does not make Chinese election interference in 2024 more or less likely, but it does very likely make any potential election interference more effective, if Beijing does decide to get involved. So far, it appears that the Chinese leadership has avoided directly interfering in U.S. presidential elections, but, as U.S.-China relations continue to worsen, such luck should not be taken for granted. Biden should make clear to Xi, publicly and privately, that any Chinese interference in the upcoming 2024 elections will be unacceptable, and met with a strong response.
China’s lie, in this case, can be the starting point of something good. Particularly if the Biden administration sees it for what it is: not just a lie, but the opportunity to begin safeguarding the U.S. 2024 elections from the growing threat of Chinese election interference.
Nathan Beauchamp-Mustafaga is a policy researcher at RAND, where he focuses on Chinese strategies for social media manipulation.
