Just months before the presidential election, Facebook says it's taking steps to combat deepfakes, videos that have been manipulated using artificial intelligence to make it appear that someone has said or done something they haven't.
Facebook is banning videos that are "edited or synthesized" to fool users but will allow parody and satire.
Why now? Many fear this insidious form of digital disinformation could be used to mislead voters, much the same way that fabricated news stories influenced public opinion in 2016.
So what are deepfakes anyway and just how worried should you be that the video you are watching isn't real?
Fake but convincing videos Deepfakes are videos doctored using cutting-edge artificial intelligence, or AI, to distort reality. The technology, which analyzes real images to generate fake ones, is a growing form of disinformation and social media platforms have been struggling with how to deal with it.
Facebook and other tech companies are sponsoring a "Deepfake Detection Challenge" to encourage AI researchers to develop new ways to automatically detect doctored videos.
Facebook disinformation:What you can do to stop its spread
Wait, is that video real? The dangers of manipulated recordings
The ability to create fake videos or manipulate existing videos has been around for decades but typically required some software or skill. In recent years, tools have become popular on social media and elsewhere that allow anyone to manipulate images and video, though this footage usually appears obviously manipulated.
Today, powerful new technologies are making it cheaper, faster and easier to produce deepfakes that are "nearly indistinguishable from reality," says Hao Li, associate professor of computer science at the University of Southern California.
There are also simpler but still effective ways to hoodwink social media users called "cheap fakes," which typically involve editing videos to spread disinformation or propaganda.
Digitally manipulated video is dangerous because we tend to have faith in what we see, Li says. "It's a real problem, and it's something that is advancing really quickly," he says. "Regulators and lawmakers are not catching up with this kind of technology."
Facebook's new deepfake policy is too narrow, says Hany Farid, a computer science professor at University of California, Berkeley specializing in digital forensics.
Under the policy, Facebook would not prohibit videos using lower-tech methods of spreading disinformation, such as last year's video of House Speaker Nancy Pelosi which was edited to make it seem as though she was slurring her speech, Monika Bickert, vice president of global policy management, told lawmakers during a Wednesday congressional hearing.
"What they are saying is: If it's a misleading video of someone saying something they didn't say and it was created using this specific type of technology, deepfakes, then we reserve the right to take it down, but if it was generated using some other kind of technique, we don't have a problem with that," Farid says. "That doesn't really make sense. The issue is not how it was made, the issue is: Is it misleading and is it harmful?"
Rachel Thomas, director of the Center for Applied Data Ethics at University of San Francisco, says she expects "cheap fakes" and the more sophisticated deepfakes to proliferate ahead of the presidential election as part of shadowy campaigns to sway public opinion.
Imagine a fake Donald Trump or a fake Joe Biden simulated saying a racial slur. Then imagine how quickly those videos would spread on Facebook or YouTube.
"The stakes are very high," Thomas says. "This kind of misinformation can have a big impact."
When used to target elected officials or political candidates, doctored video can erode trust. Nearly two-thirds of Americans recently surveyed by Pew Research said altered images and videos caused a great deal of confusion in understanding the basic facts of current events.
The term "deepfakes" refers to videos that have used artificial intelligence techniques to combine and superimpose multiple images or videos onto source material. How can you spot deepfakes? How can you tell if the video you're watching is a deepfake?
"That is becoming incredibly difficult. The technology behind fake videos is improving almost on a continuous basis," says Siwei Lyu, computer science professor at the State University of New York at Albany and a member of the Deepfake Detection Challenge's advisory group.
He recommends "psychological preparedness," as in being constantly on the alert for videos, images and audio that have been altered.
"Whenever you see an interesting video showing something that is bizarre or exceptional, a certain vigilance should be raised," he says.
Examine the video carefully before sharing it on social media. Is the video low-resolution or grainy? Is it a single person talking in the video? Is it relatively short, say 30 seconds or 60 seconds long?
Is the lighting strange or the face discolored or blurry? Is there blurriness between the face and neck or between the face and the hair? Is the sound not synced with the images?
Some of the other tell-tale signs discernible to the naked eye, according to Subbarao Kambhampati, a computer science professor at Arizona State University: different-sized eyes or ill-formed teeth, or more than two eyes, or inconsistencies in the background of the video.
Who inspires you? USA TODAY seeks your Women of the Century to commemorate 19th Amendment Social media is full of lies, and the Iran attacks show how dangerous they've become Takata recalls 10 million more airbags from 14 automakers The Daily Money: Subscribe to our newsletter But, says Kambhampati, the rapid improvements in deepfake technology means that we will soon have to rely on AI techniques to detect what the human eye cannot.
"There is not a 100% foolproof way of identifying deepfakes, not even for AI researchers," Thomas says. "Detection is always going to be an arms race. As people develop more accurate detection algorithms, fakers will develop even more sophisticated frauds."
There are non-technical ways to sniff out a deepfake, just like other forms of disinformation. Ask yourself: Who is the person publishing this information? Is this person reliable? What else has this person posted? Are the claims in the post backed up by sources you trust?
When in doubt, turn to nonpartisan fact-checkers who've spent hours, sometimes days, tracking down accurate information, Farid says.
"You never want to jump to a conclusion one way or another, you never want to say: I looked at this and I'm sure it's real or I looked at this and I'm sure it's fake," he says. "The best you can really hope to do is to say: You know what? I should be careful here because I'm not sure, and then research it."
Fair warning: It's going to get worse Li says his company, Pinscreen, will demonstrate for attendees of the World Economic Forum in Davos how to livestream fake video.
He says it's part of an effort to raise awareness of the need for technologies that root out deepfakes.
"In real-time, you can reenact yourself as Will Smith or some celebrity or a politician. You can be George Bush," he says. "The results are very convincing."
Link to Story Source
Image Search: | Last 48 Hours | Last 30 Days | All Time
Story Search: | Last 48 Hours | Last 30 Days | All Time