'Deep fake' videos that can make anyone say anything worry U.S. intelligence agencies | What Is IT?
NEW YORK - A video of a seemingly real news anchor, reading a patently false script saying things like the "subways always run on time" and "New York City pizza is definitely not as good as Chicago" gives a whole new meaning to the term fake news.
But that fake news anchor is a real example of a fascinating new technology with frightening potential uses.
I was stunned watching the Frankenstein mix of Steve Lacy's voice coming out of what looks like my mouth.
"That's how well the algorithm knows your face," Professor Siwei Lyu told me.
The video is what is known as a deep fake: a computer-generated clip using an algorithm that learned my face so well that is can recreate it with remarkable accuracy.
My generated face can be swapped onto someone else's head (like that original video with Steve) or it can be used to make me look like I'm saying things I've never said.
For this piece, I worked with Lyu and his team at the College of Engineering and Applied Sciences at the University at Albany.
For many people, seeing is believing.
"I would say it's not 100% true anymore," Lyu said.
Their deep fake research is funded by the Defense Advanced Research Projects Agency, or DARPA, which acts as the research and development wing of the U.S. Defense Department. They're working to develop a set of tools the government and public can use to detect and combat the rise of deep fakes.
"What we're doing here is providing a kind of detection method to authenticate these videos," Lyu said.
What's more, deep fakes technically aren't that hard to make. All it takes is a few seconds of video of someone, a powerful computer, and some code, which Lyu and his team don't release publicly.
"The real danger, I believe, is the fact that the line between what is real and what is fake is blurred because of the existence of this kind of technology," Lyu said.
But it is about more than just a news anchor face-swap experiment. The power to make a video of anybody saying anything is alarming.
Even the former president is raising red flags. The funny thing is (as you see in the video) that is not Barack Obama. The video is a deep fake. Actor Jordan Peele is impersonating Obama's voice. The algorithm is doing the rest. It's meant to be a PSA about the dangers of deep fakes.
"Moving forward we need to be more vigilant with what we trust from the internet," fake Obama warns.
Imagining how a deep fake video could quickly create a very scary real-world scenario is not hard.
Say, for instance, a video of a world leader, such as Vladimir Putin, pops up on the internet declaring war on another country, or, maybe, the head of a major company announcing his or her abrupt resignation, putting the markets in a tail spin.
Videos like that can spread like wildfire before fact checkers, journalists, and governments even have the chance to authenticate it.
And the U.S. government is paying attention. Deep fakes were a topic at the recent worldwide threats hearing in front of the Senate Intelligence Committee.
"Are we organized in a way where we could possibly respond fast enough to a catastrophic deep fakes attack?" Sen. Ben Sasse, a Nebraska Republican, asked a panel of the heads of the nation's intelligence agencies.
Director of National Intelligence Dan Coats responded by saying emerging technology like deep fakes pose "a major threat to the United States and it's something the intelligence community needs to be restructured to address."
House Intelligence Committee member Sean Patrick Maloney told Fox 5 News, "You ain't seen nothing yet."
"Because when you start to get a look at these deep fake videos, which are compelling but false, it's an order of magnitude more serious than anything we've seen to date," Maloney added. "That's why it's important we get there first, deter the bad guys, and ask the private sector to step up."
That's where Lyu's research comes in. His main focus is actually detecting and preventing deep fakes.
"It's a cat-and-mouse kind of game. Each side wants to get a little bit of an edge over the other," he said. "And this actually, for the good part, gives us motivation and incentive to grow this research field."
But as the research grows, so will the quality of the deep fake videos being produced, which will make differentiating between what's real and what isn't increasingly difficult. That is why they're working to develop this set of tools right now to be deployed down the road.
"Our challenge is: how do you build the algorithm to identify the anomaly?" Lt. Gen. Robert Ashley, the director of the Defense Intelligence Agency, offered at that recent Senate threats hearing. "Because every deep fake has a flaw. At least now they do."