China’s Cyberspace Administration has issued guidelines on how to do deepfakes the right way.
Deepfakes use artificial intelligence to create realistic depictions – usually videos – of humans saying and/or doing things they didn’t say and/or do. They’re controversial outside China for their potential to mislead audiences and create trouble for the people depicted.
Beijing clearly also has worries about the technique as the Cyberspace Administration (CAC) has issued regulations that prohibit their creation without the subject’s permission, or to depict or utter anything that could be considered as counter to the national interest. Anything counter to socialist values falls under that description, as does any form of “Illegal and harmful information” or using AI-generated humans in an attempt to deceive or slander.
But the rules also suggest China expects synthetic humans will be widely used. For instance, they allow use of deepfakes in applications such as chatbots. In such scenarios, deepfakes must be flagged as digital creations.
The document also envisages that deepfakes will be used by online publishers, which must take into account China’s myriad other rules about acceptable online content.
Including the one that censpored images of Winnie the Pooh online, as the beloved bear – as depicted by illustrator E. H. Shepard – was felt to resemble, and mock, China’s president-for-probably-life Xi Jinping.
The Register therefore suggests it will be a very, very, brave Chinese developer that creates a photorealistic ursine chatbot or avatar.
The regulations also spell out how the creators of deepfakes – who are termed “deep synthesis service providers” – must take care that their AI/ML models and algorithms are accurate and regularly revised, and ensure the security of data they collect.
The rules also include a requirement for registration of users – including their real names. Because allowing an unknown person to mess with deepfakes would not do.
The rules are pitched as ensuring that synthesis tech avoids the downsides and delivers benefits to China. Or, as Beijing puts it (albeit in translation), deepfakes must “Promote the healthy development of internet information services and maintain a good ecology of cyberspace.”
The regulations come into force on January 10, 2023. ®