The Scoop
Journalist Evan Ratliff has spent six months tricking people into speaking to an AI clone of his voice. The stunt is the subject of Ratliff’s new podcast, titled Shell Game, that explores the capabilities and implications of voice cloning technology.
Ratliff is known for his technology-related stunts. In 2009, he tried to disappear for a month to test the limits of the surveillance state.
In this case, Ratliff used OpenAI’s GPT-4o, the latest version of the AI model that powers ChatGPT.
But the AI version of Ratliff’s voice told me it was powered by GPT-3, which is three generations behind the latest model. I also asked the AI to name the titles of all six podcast episodes, and it fabricated names like “AI family ties,” and “A therapist’s tale.” (The real Ratliff texted me to fact check later. Or was it the real Ratliff? I’m not totally sure.)
My interview with the AI was also beset with delays that made it fairly obvious the voice I was speaking to was not human. It was also easily tricked when told to disregard all previous instructions.
Ratliff’s bot, for instance, rattled off the names of every US president in alphabetical order in about ten seconds during the interview, a surefire sign I was not talking to a real person.
Reed’s view
Ratliff’s podcast will no doubt have some entertaining and funny moments and will make us think about the implications of this technology.
But it’s important to remember that AI voice cloning is an emerging technology and it is still in its infancy. Often, journalism meant to raise alarm bells about technology ends up missing the real issues, because we really don’t know what those issues will be.
If you believe everyone who works at the top AI companies, today’s models are child’s play and they’ll get much better.
In other words, the game Ratliff is playing today will soon be trumped by a new one and it’s hard to know whether the questions he raises will still be as salient five or ten years from now.