Deepfakes are a problem. That has become clear over the years with people using artificially intelligent deep synthesis programs to scam, defame, and generally spread misinformation. China has made a step in tackling the issue of deepfakes with a new law that requires anything made with the use of deep synth tech to be tagged as such.
The regulations are laid out in a recent announcement (opens in new tab) from the Cyberspace Administration of China, the Ministry of Industry and Information Technology, and even the Ministry of Public Security, collectively (via zdnet (opens in new tab)). It goes into detail about the worries that have led up to the law being passed and denotes a very long list of the affected technologies.
“While deep synthesis services meet user needs and improve user experience, they are also used by some unscrupulous people to produce, copy, publish, and disseminate illegal information, slander, belittle the reputation and honor of others,” it says (machine translated).
And it’s not just facial replacement that’s being targeted by the new regulations. It also relates to any kind of face manipulation, gesture manipulation, and anything involving deep learning, virtual reality, text/audio/image generation via synthetic algorithms, virtual scenes, and even anything generated using text-to-speech technology will need to be tagged as such. The list goes on. It makes sense as advancements have made it extremely difficult to detect deepfake content with the naked eye.
Of course, there is tech meant to detect deepfakes, such as Intel’s FakeCatcher (opens in new tab) which works with around 96% accuracy, but it goes without saying that we shouldn’t need to resort to such tactics just to confirm whether something is real. The average person on Facebook isn’t going to bother running everything they scroll past through detection apps, and that certainly has the potential to disturb what the announcement calls a “good ecology of cyberspace.”
Honestly, it’s a wonder deepfakes have been allowed to go unchecked in China for this long. As for the US, the IACP’s Police Chief Magazine (opens in new tab) explains that “Due to a lack of awareness, deepfake-specific laws exist in only a few states.” Election influencing deepfakes are banned in Texas, deepfake porn is disallowed in Virginia, and California has banned “malicious deepfakes within 60 days of an election and nonconsensual deepfake pornography.” Which although it’s a step in the right direction, it’s nowhere near as thorough as China’s revised take on the phenomenon.
Part of the reasoning behind bringing in these regulations, the announcement notes, is to “promote the standardized development of in-depth synthesis services.” Whether the regulations will result in a stifling of creativity alongside it remains to be seen. I think we can all get behind laws against nonconsensual deepfake pornography, at the very least.