AI-powered deepfake video technology is advancing faster than what even some of the world’s top experts thought was possible.
Its stunning trajectory means that within a few years, perfect and virtually undetectable deepfakes may be possible, USC professor and deepfake pioneer Hao Li said today at MIT Technology Review’s EmTech conference.
Li is collaborating closely with Hany Farid of UC Berkeley, one of the world’s best digital forensics scientists. The collaboration reflects both competition—after all, Farid’s job is to detect the kind of deepfakes Li creates—and, in a bigger sense, close cooperation.
“We are working together on an approach that assumes that deepfakes will be perfect,” Li said. “Our guess that in two to three years, it’s going to be perfect. There will be no way to tell if it’s real or not, so we have to take a different approach.”
Li also runs a startup called Pinscreen focused on business and entertainment uses of the tech. His team was behind the wizadry that put Paul Walker into Furious 7, part of the Fast and Furious film franchise, after Walker died in a car crash. In a sign that Hollywood’s knocks are getting louder, he’s currently working on a Will Smith movie project. But Li has complex feelings about the future of the technology he’s working so closely on.
“One theory on why it hasn’t caused even more harm yet is simply because the quality isn’t there yet,” Li said. “But it will get there. And a second theory is that the technology still requires a little bit of work from users to create really good deepfakes. But it’s funny because a few months ago I said it may be years until that moment where it’s easy for anyone to do this. No one realized there was this Chinese app that came out a week later. I thought, ‘Wow, this is developing more rapidly than I thought.’”
Real Life. Real News. Real Voices
Help us tell more of the stories that matterBecome a founding member
The app is Zao, a face-swap app that, with a single photograph, can let you place yourself in big television shows and movies. The app, which is monstrously popular in China, is at the center of a debate about privacy and unintended consequences.
In case you haven’t heard, #ZAO is a Chinese app which completely blew up since Friday. Best application of ‘Deepfake’-style AI facial replacement I’ve ever seen.
Here’s an example of me as DiCaprio (generated in under 8 secs from that one photo in the thumbnail) 🤯 pic.twitter.com/1RpnJJ3wgT
— Allan Xia (@AllanXia) September 1, 2019
It’s a powerful tool that shot up to the top of China’s App Store for obvious reasons. Who doesn’t want to instantly become Leonardo DiCaprio?
Is it perfect? Far from it. But the fact that so much can be done within a few seconds on anyone’s smartphone signals just how fast deepfake technology is advancing—and just how many crucial questions remain unanswered.
Subscribe to the newsletter news
We hate SPAM and promise to keep your email address safe