Race to defuse deepfake videos: UC Berkeley researchers creating software for newsrooms

Kate Larsen Image
ByKate Larsen KGO logo
Monday, July 8, 2019
UC Berkeley researchers creating software to diffuse deepfake videos
There's a race to diffuse a powerful political weapon: computer-generated fake videos. Researchers at UC Berkeley are leading the charge to fight the fake news technology.

BERKELEY, Calif. (KGO) -- There's a race to defuse a powerful political weapon: computer-generated fake videos. Researchers at UC Berkeley are leading the charge to fight the fake news technology.

Millions of people saw distorted video of House Speaker Nancy Pelosi, posted to social media in May. Simple edits to the video, made it sound like Pelosi was drunkenly slurring her words, when she was not.

"That really wreaks havoc on democracy, on society and our personal safety," said Hany Farid, a Computer Science Professor at UC Berkeley.

RELATED: Altered videos of Speaker Nancy Pelosi slurring words goes viral

Farid and graduate student, Shruti Agarwal, are developing software to combat deepfake technology -- content synthesized through artificial intelligence.

An example of a deepfake video, can be seen in filmmaker Jordan Peele's PSA from 2018 about the danger of the videos.

"This technology allows you to create highly sophisticated fake content and it doesn't require a lot of skill to create the content," explained Farid.

"We started tracking when a person is talking, how their eyebrows are moving, how their lips are moving, how their cheeks are moving," said Agarwal, as she demonstrated how her software works.

Ahead of the 2020 election, Agarwal is mapping how specific politicians and their faces move when they speak, a reference point against any manipulations.

By the end of the year, Farid and Agarwal want the technology to be ready for newsrooms, so that if there's a question about the authenticity of a video, newsrooms can run it through their software, via a UC Berkeley portal.

Farid says her program will allow them to determine the likelihood of altered videos, ultimately deterring bad actors. "We really want our technology, so that the creation of a compelling fake, becomes so difficult that an average person can not create it so easily."

"The constitution gives you the right to say anything that you want to, you have the freedom of your speech. But the constitution doesn't give you the right to put your words into my mouth and that's what a lot of this technology does," said Palo Alto Assemblyman, Marc Berman.

Berman has proposed a law, CA AB730, which would make it illegal to create or disseminate manipulated audio, video, or images of a candidate for office, 60 days before the campaign. Unless Berman says, there's a disclaimer label on the content.

"You can create these videos if you want to, but you have to be honest with the public about the fact that it's fake."

Berman hopes his bill will be signed into law in January 2020.