Login et outils

Tech’s deepfake problem is worsening

 — 4 septembre 2020
This week, thousands of Twitter and Facebook users, including White House Deputy Chief of Staff Dan Scavino, circulated a manipulated video appearing to show Democratic presidential candidate Joe Biden falling asleep during a live interview. The run-up to the U.S. presidential election is also speeding up the arrival of a tipping point for digital fakery in politics.

Experts are worried about Silicon Valley’s ability to meet the challenge once AI-generated deepfakes become widespread and it’s easy to make any famous person appear to say anything.

Some especially tricky challenges we face or will soon face, according to experts:

  • « Cheapfakes, » or « shallow fakes » — like the slowed-down video of Nancy Pelosi — can spread quickly before getting caught, and even then may not be taken down.
  • « Readfakes » are on the rise — a coinage from Graphika researcher Camille François referring to AI-generated text, which can take the form of fake articles and op-eds.
  • Generative Adversarial Networks, which can create images of nonexistent people, let disinformation campaigns make fake social media accounts or even infiltrate traditional media.
  • « Digital humans » expand on that idea, relying on voice synthesis and faked video to create entire faked personas.
  • Sheer volume is a concern, as AI gets better at generating a lot of images or text at once, flooding the internet with junk and making people less sure of what’s real.
  • Those sharing faked media are getting smarter about skirting the line of breaking platform rules, such as by claiming a video is just a parody. Of course, some manipulated videos are meant as parodies, which only makes this problem tougher.

Solutions are tough to come by. Experts agree that attempting to catch and kill deepfakes and cheapfakes on a case-by-case basis may never work at scale. But some tactics can help.

  • Putting deepfake detection tools in users’ hands would help platforms address the challenge of scale, François told Axios.
  • Best practices shared across the industry are a must, said Rob Meadows, chief technology officer at the AI Foundation, which recently partnered with Microsoft on Reality Defender, a deepfake detection initiative.
  • Authenticating images and videos by tracking when they’re taken and every subsequent edit or manipulation could prove more effective in restoring trust than trying to detect and quash deepfakes once they’re already circulating.


Laissez vous dériver… choisissez votre prochaine étape

A global atomic renaissance
Worldbuilding – Gaming and Art in the Digital Age
Dawn, Avatar Robot Café
(Im)possible Baby
Agriculture de précision : pourquoi est-ce une fausse bonne idée ?
Des scientifiques créent des bâtiments imprimés en 3D en terre
Unearthing the Secret Superpowers of Fungus
Concept art of the Yuanmeng
L'Allemagne inaugure la première flotte de trains à hydrogène au monde
Designer materials to keep plastic out of landfills