Deepfakes are an increasingly popular editing technique that uses artificial intelligence to alter faces and voicesin videos. Part of the name comes from the AI used to create them, deep learning. Similar technology has been used in Hollywood for years, but with AI getting better and more accessible, today they’re all over the internet.
Deepfake Videos Are Getting Terrifyingly Real
Published: April 5, 2019
Onscreen: All of these videos...are fake. An increasingly popular editing technique uses artificial intelligence to alter faces and voices. The end product is called a "deep fake."
Deepfake Obama: You see, I would never say these things … at least not in a public address. But someone else would. Someone, like Jordan Peele.
Onscreen: Part of the name comes from the AI used to create them...deep learning. And they're all over the internet.
Collin Frend: It’s a lot more research than you would think that goes into making a goofy video like that. I didn’t do any after-touching, that was just using the technology available from the machine learning side.It’s sort of learning how to re-create a person’s face from looking at thousands of images over and over and over again.
Onscreen: Some from of this technology has been used in Hollywood for years, but the AI is getting better and more accessible.
Hany Farid: You don’t have to know what Github is, you don’t have to know how to program in Python — none of that matters. You pay somebody 20 bucks and they’ll create the fake for you. That’s a bit of a game changer.
Onscreen: But deep fakes of average people could be really damaging.
Danielle Citron: For the everyday person who a deepfake sex video emerges in a Google search of your name, for that person it can be almost impossible to debunk.
Onscreen: Some worry about the impact a convincing deep fake might have if it's strategically timed...like on the eve of the 2020 election.
Farid: While I don’t think that’s likely, I don’t think it’s out of the question. And that’s enough to keep me up at night. You don’t have to fool tens of millions of people. The last national election here in the US was 80,000 votes in eight states. The margins are very very thin.
Tim Hwang: I don’t think that a deep fake will be a major cause of disruption during the campaign. Think about a well-placed rumor before an IPO. I’m not sure how much machine learning is adding to this. I find it very hard to believe that there is something uniquely devastating about video.
Onscreen: The state department is working with experts to identify and combat malicious deep fakes.
Farid: Right now we are developing techniques to protect world leaders from deepfakes we are particularly worried about how the Donald Trump, the Theresa May’s of the world, how their likeness will be used to disrupt elections, incite violence by creating fake content. There is a weaponization of the technology and I am simply advocating that we as technologists starting thinking, not after the fact, but before the fact about how we can start putting some safeguards in place.
Digital Producer: Emily Zendt
Production Assistance: Rishya Narayanan, Taylor White
Additional Footage: University of Washington, ACM Publications, University at Albany, SUNY, Collin Frend, Caroline Chan, MIT
© WGBH Educational Foundation 2019