they keep coming. First, there were the pixelated likenesses of actresses Gal Gadot and Scarlett Johansson brushstroked into dodgy user-generated adult films. Then a disembodied digital Barack Obama and Donald Trump appeared in clips they never agreed to, saying things the real Obama and Trump never said. And in June, a machine-learning-generated version of Facebook CEO Mark Zuckerberg making scary comments about privacy went viral.
There’s concern the fallout from the technology will go beyond the creepy, especially if it falls into the hands of rogue actors looking to disrupt elections and tank the shares of public companies. The tension is boiling over. Lawmakers want to ban deepfakes. Big Tech believes its engineers will develop a fix. Meanwhile, the researchers, academics, and digital rights activists on the front lines bemoan that they’re ill equipped to fight this battle.
Witness has been training media companies and activists in how to identify A.I.-generated “synthetic media,” such as deepfakes and facial reenactments—the recording and transferring of facial expressions from one person to another—that could undermine trust in their work. He and others have begun to call on tech companies to do more to police these fabrications. “As companies release products that enable creation, they should release products that enable detection as well,” says Gregory.