Beijing announced on Thursday that it has summoned eleven internet companies, including Alibaba, Tencent and ByteDance, over "audio-based social networking apps and apps that involve 'deepfake' technologies."
The announcement from China's Cyberspace Administration and Ministry of Public Security came as a direct response to an obsession among urban elites with Clubhouse-like apps and the rapid proliferation of viral DeepFakes in China. Protocol has covered the rise of the Clubhouse clones here. But what's going on with DeepFakes? Here's what you should know:
- Thank Russia for China's deepfake craze. At the end of February, Avatarify, an AI-based app developed by Russian engineers that transform static portrait photos into videos, trended on Douyin, the Chinese version of TikTok. Chinese users were quick to find innovative ways to create hilarious videos with the app, including one showing Elon Musk and Jack Ma seemingly singing the popular song Dragostea Din Tei in sync. See for yourself
Clocwise, from top left: Elon Musk, Mark Zuckerberg, Warren Buffet, Jack Ma, Wang Jianlin, Jia Yueting, Lei Jun, Pony Ma.Protocol screengrab/Avatarify
- Deepfakes have gotten very popular in China very quickly. Over 7 million Douyin users have used similar filters to create videos, and the videos have been watched over 2.7 billion times combined, according to the Chinese outlet TechSina. Taobao vendors started charging to create deepfakes, to people who don't have iPhones and thus can't use Avatarify themselves. On February 28, Avatarify topped the Chinese App Store chart; just two day later, it was pulled for reasons that were not made public.
- The videos that trended this time were mostly benign, where ordinary users or even celebrities use their own photos for comedic effects. But AI face-swapping technologies have previously been used for creating pornographic content and disinformation. Academic researchers have been cautious in developing the commercial application of deepfake technologies, for this very matter. And no regulator wants to wait until there's a critical or incriminating deepfake of a Politburo member or Xi Jinping before taking action.
- The summons is essentially Beijing asking tech companies to conduct self-censorship more quickly and thoroughly. The announcement cites a 2018 regulation that required Internet companies to conduct security assessments of their own apps and services when "online services that involve public opinionating or public organizing are introduced," or when "the number of users experiences a significant growth."
- China's now one of the world's deepfake capitals. According to a deepfake white paper published by Nisos, a cybersecurity intelligence firm, Russia, Japan and China are the three most prominent countries where deepfake creators reside. In China, "the typical deep fake purveyor profile is more likely the hobbyist and researcher," the whitepaper says.
- Chinese startups are trying to find a profit model for deepfake-like technologies. Surreal, a one-year-old company backed by Sequoia and ZhenFund, said it wanted to become "the Google Translate of videos." It can modify people's facial characteristics and change the language they speak in videos, therefore reducing the cost of international marketing, the founder told Chinese tech publication 36Kr.