AI can hijack anyone's identity
The Horror of Deepfake Nudes
Written for UnHerd
Updated Mar. 5, 2021
Published Oct. 28, 2020
Non-consensual porn isn’t a ‘woman’s issue’ — it shows that anyone’s identity can be hijacked.
Martin Scorcese’s most recent film, The Irishman, told a story that spanned seven decades. Robert Di Niro and Joe Pesci starred, and in order to “de-age” them, Scorcese used a special three-rig camera and employed dedicated special effects artists for post-production work. The costs ran into the millions — and the results were patchy. Earlier this year, a YouTuber decided to see if he could do any better: using free artificial intelligence software, he bettered Scorcese’s attempt in a week.
It is no exaggeration to say that soon almost everything we see or hear online will be synthetic — that is, generated or manipulated by AI. Machines that can “learn” to do almost anything when ‘trained’ on the right data — and they’ve never had access to more data, nor so much power to churn through it all. Some experts estimate that within 5-7 years, 90% of all video content online will be synthetic. Before long, anyone with a smartphone will be able to make Hollywood level AI-generated content.
One synthetic-text generating model can already generate articles that appear to have been written by a human. AI can be trained to clone someone’s voice even if they’re already dead: an old recording of JFK’s voice has been used to make a clip of the former president reading the Book of Genesis. AI trained on a dataset of human faces can generate convincing fake images of people who do not exist, and it can be taught to insert people into photographs and videos they were not originally in. One YouTuber is working on a project to insert actor Nicholas Cage into every movie ever made.