“
One piece of AI, called the generator, is instructed to create a deepfake showing, for example, Hillary Clinton endorsing Ron DeSantis. The generator is provided with sufficient raw data, including video footage and voice recordings of Clinton. It then uses the data to create an initial video.
That video is passed to a different piece of AI, called the discriminator. The discriminator’s job is to sniff out counterfeits. When it looks at the generator’s first draft of the video, it can tell it’s a fake. So, the discriminator passes it back to the generator and basically says, “Fake news!”
The generator looks at what tipped the discriminator off that the video was counterfeit, makes some changes to address those issues, and then sends it back to the discriminator for another evaluation. It fails the “sniff test” again, and the clip goes back to the generator for a third iteration.
Rinse and repeat a million times or more. Finally, maybe on version 1,438,847, the discriminator looks at the video and says, in effect, “Holy cow! Hillary Clinton endorsed Ron DeSantis!
”
”
Craig Huey (The Great Deception: 10 Shocking Dangers and the Blueprint for Rescuing The American Dream)