How Faking Videos Became Easy and Why That's So Scary

A minute-long video of Barack Obama has been seen more than 4.8 million times since April. It shows the former U.S. president seated, with the American flag in the background, speaking directly to the viewer and using an obscenity to refer to his successor, Donald Trump. Or rather, his lips move as the words are spoken. The video is actually a so-called deep fake made by actor-director Jordan Peele, who impersonated Obama’s voice. Peele created the video to illustrate the dangers of fabricated audio and video content depicting people saying or doing things they never actually said or did. Researchers at New York University describe deep fakes as a “menace on the horizon.”

While manipulation of digital files is nothing new, this breed of believable fakery is accomplished using computer programs that employ a form of artificial intelligence. An algorithm is trained to recognize patterns in actual audio or visual recordings of a particular person, a process known as deep learning. As with doctored images, a piece of content can be altered by swapping in a new element -- such as someone else’s face or voice -- and seamlessly joining the two. The manipulations are most misleading when combined with voice-cloning technology, which breaks down an audio recording into half-syllable chunks that can be reassembled into new words -- the same method that’s used to create voice assistants like Apple’s Siri and Amazon’s Alexa.