Detecting Deepfakes: Inside the growing threat of AI generated videos

News

CLEVELAND (WJW) — Say you’re scrolling through your social media and come across a video that looks like a trusted a local newscast.

“Good evening, this just in,” the anchor says. “There will be no snow or cold weather in Cleveland this winter. The Browns, Indians and Cavs are all in first place and on the verge of winning championships, and everyone is getting a pay raise.”

But something doesn’t seem right, the anchor looks like me, Joe Toohey, but sounds like FOX 8’s Gabe Spiegel.

“We took a pair of videos: one video from you and one from your colleague,” Professor Siwei Lyu tells me.

That video is what’s known as a “Deepfake.” It’s a fascinating technology with frightening potential impacts. It’s a computer generated clip made using an algorithm which was trained using images of my face.

It learned it so well that it can recreate it to make it look like I’m doing and saying things I’ve never done.

“The way to train this is really like teaching a baby to recognize human faces and then hallucinate or dream about human faces,” Professor Lyu says. “So give this model tons and tons of images of human faces.”

“The model will get better and better as it’s seeing more data and then it will create images that look more and more realistic.”

For this piece, FOX 8 anchor Joe Toohey worked with Professor Siwei Lyu and his team at the University at Buffalo. Some of their research is funded by the Defense Advanced Research Projects Agency (DARPA).

For many, seeing and hearing is believing.

Because of Deepfakes, Lyu says, “we need to take a lot of care with what we see and what we hear.”

“Seeing is not believing. Hearing is not believing either.”

The technology has countless potential uses. Deepfakes of National Football League stars Baker Mayfield and Saquon Barkley appear in an ad for Hulu.

There’s also a darker, more dangerous use case. The scenario isn’t hard to conjure up: imagine a video of a world leader like Vladimir Putin appears on the internet saying he’s about to invade another country.

The clip could travel the globe online before governments, journalists, and fact checkers ever have a chance to authenticate it.

Senator Rob Portman (R-OH) says he was “shocked” when he first learned about Deepfakes.

He then co-founded the U.S. Senate’s Artificial Intelligence caucus. It’s a bipartisan group of senators pushing legislation to get the government up to speed on Deepfakes.

“We’ve got actually three more bills we hope to pass in the next month or so that try to get the federal government more involved in understanding what Deepfakes are and coming up with some rules around them,” Portman says. “Because it really is pretty scary.”

Portman says the threat from Deepfakes and manipulated media requires a “whole of government” approach. One of the bills the AI Caucus is pushing would create a new body within the government as a “center of excellence” to handle the growing Deepfake problem.

“In the absence of our legislation passing, I’d have to give the federal government a pretty low grade because we don’t have a centralized way to deal with this and understand it,” Portman tells FOX 8.

“We don’t have the expertise we should have. Government’s a little behind generally when it comes to technology. And I think we have to catch up — artificial intelligence generally, Deepfakes in particular — and my hope is that we will.”

Professor Lyu is working on just that: detecting and preventing the spread of Deepfakes.

Whether through his new website “Deepfake-o-meter”, which allows users to submit media to analyze its authenticity or preventing them at the start by building a sort of digital watermark for people’s private data.

He’s developing methods to be used now and compares it to medicine used to help stop the spread of a Deepfake virus.

“I gave a talk at a senior center and one gentleman after my talk came to me and said, ‘that was scary. I will never trust anything I see on the internet.’”

“But that’s the fear I have,” Lyu says. “Because if we’re not trusting any information, we’re blocking out useful information.”

Get the latest headlines from FOX8.com below:

FOX 8 Cleveland Weather // Quick Links:

Hot on FOX 8

More Viral

continue reading override