Facebook Twitter

Researchers from Berkley and USC have developed an AI that can detect deepfakes

SHARE Researchers from Berkley and USC have developed an AI that can detect deepfakes

SALT LAKE CITY — Deepfakes can be difficult to detect, but a new tool can identify subtle differences that mark the artificial videos as fakes.

Gizmodo reports that researchers from UC Berkeley and the University of Southern California have developed artificial intelligence that identifies “soft biometric” signatures. This means the program looks for subtle hand, head, eye and body movements that make each of us unique.

Interestingly enough, many of these tics aren’t even perceptible to the naked eye — but when they’re not there, it’s noticeable.

“It’s all done subconsciously—you don’t realize your body is doing it, nor does your mind immediately recognize when someone else is — but as a result, it’s a detail that current deepfake processing techniques don’t take into account when creating a fake,” Gizmodo notes.

According to the research team’s findings, the new AI — which employs techniques used to create deepfakes in the first place — was able to identify deepfakes 92 percent of the time.

Ideally, the AI could be used to detect fake videos that could sway public opinion in the upcoming 2020 primary elections, assuming it can keep up with evolving deepfake techniques.

What are deepfakes?

According to Mashable, deepfakes are face-swapping videos that can overlay one face on top of another, creating a fake video. Vox reports that AI-generated photos have also evolved enough to pass as real people.

Deseret News has also reported that easy access to convincing deepfakes could be used to target government officials, agencies and institutions to undermine American democracy and morale.

However, AIs and deepfakes can be used for less-insidious purposes. I’ve written for Deseret News that deepfake techniques can be used to make older photos and paintings — like the Mona Lisa — smile.