Sep 19, 2019

Bogus Putin interview reveals a real deepfake threat

The real Putin. Photo: Kirill Kudryavtsev/AFP/Getty

At an MIT conference on Wednesday, a journalist pointedly asked Russian President Vladimir Putin whether he would interfere again in U.S. elections. Putin demurred.

What's happening: The world leader was actually a glitchy deepfake. His face was a real-time AI-generated mask that made a performer look like Putin on screen — but because the mask stopped at the forehead, this was Putin with a fresh head of hair.

The big picture: The stunt was a snapshot of the current state of deepfakes, a fast-improving technology that has stirred up concern for its potential to disrupt elections and business — but that also holds creative promise.

  • For the most part, deepfakes require a considerable amount of time, setup and fine-tuning — they're generally prerecorded video clips.
  • But the Putin fake was different. It masked the speaker's face in real time — but as a result, it was considerably less convincing than the typically better-rendered clips.

Details: Putin's Wednesday cameo — portrayed by MIT Technology Review editor-in-chief Gideon Lichfield, who also played himself as the interviewer — was the work of Hao Li, a USC professor and founder of a deepfake-producing startup called Pinscreen.

  • Li's on-the-fly deepfake, which appeared on screen before a live audience, wouldn't fool anyone. There was a clear dividing line between Putin's face and Lichfield's forehead, and the faux president's mouth — a notoriously difficult element to get right in deepfake videos — often looked weird.
  • But Li has Hollywood-grade work under his belt: Pinscreen helped render the late Paul Walker in "Furious 7," and Li is reportedly working on a new Will Smith movie.

In an interview with MIT Tech Review, Li says the technology is only getting better. “Our guess that in 2 to 3 years, it’s going to be perfect," he told reporter Patrick Howell O’Neill. "There will be no way to tell if it’s real or not, so we have to take a different approach.”

  • Li is working with Berkeley's Hany Farid, a top digital forensics expert, to improve deepfake-detection algorithms. Facebook, where Farid consults, announced this month that it's pouring money into a contest to make better detectors.
  • Yes, but: That's not an easy goal. Experts say that deepfake generation and detection will be forever chasing each other — and that deepfake-makers will always have the edge.

Go deeper

Revenge of the deepfake detectives

Illustration: Sarah Grillo/Axios

Tech giants, startups and academic labs are pumping out datasets and detectors in hopes of jump-starting the effort to create an automated system that can separate real videos, images and voice recordings from AI forgeries.

Why it matters: Algorithms that try to detect deepfakes lag behind the technology that creates them — a worrying imbalance given the technology's potential to stir chaos in an election or an IPO.

Go deeperArrowSep 28, 2019

The deepfake threat to evidence

Illustration: Rebecca Zisser/Axios

As deepfakes become more convincing and people are increasingly aware of them, the realistic AI-generated videos, images and audio threaten to disrupt crucial evidence at the center of the legal system.

Why it matters: Leaning on key videos in a court case — like a smartphone recording of a police shooting, for example — could become more difficult if jurors are more suspicious of them by default, or if lawyers call them into question by raising the possibility that they are deepfakes.

Go deeperArrowOct 12, 2019

Podcast: Groupon founder's second act

Six years after being fired as CEO of Groupon, Andrew Mason has founded a podcast production startup called Descript that basically works like an audio word processing platform. But it also lets users add words they never spoke, thus raising security and deepfake concerns.

Keep ReadingArrowSep 19, 2019