University of Washington researchers last month formally unveiled new video editing technology that can digitally put words into people’s mouths. As a preview of their invention, the group released a clip of President Obama with the audio from several different speeches lip-synced over separate video footage of a single speech. The result is a believable fraud that would convince most people the former president was saying something at a time and place when he in fact said something else.
If you’re like me, hearing this will make you worry about people who might try to get the president to say something he’d never uttered before. Something sinister. However, the researchers have said their new tool will have applications in video conferencing, and that’s where the real danger lies.
With a high-tech video call editing tool, it isn’t the fake president you have to worry about, it’s the fake colleague on the other end of your workplace video conference who could really cause you trouble.
How to Spot a Fake President
If you look very closely at the University of Washington’s phoney Obama for even just a few seconds you’ll convince yourself you can see through the illusion.
Every once in awhile his lips and words seems a bit out of sync, and maybe you see a little fuzziness around the chin. That’s if you’re looking for a fake, though. At first glance you were fooled. Admit it.
The science behind the illusion laid bare at the 2017 Siggraph Conference in August, and the Obama cut and paste job really doesn’t seem to do it justice.
The new software is an algorithm that turns audio into lip-synced video of the speaker. The researchers chose President Obama as a subject because their machine-learning tech needs to analyze lots of footage of a person before it can accurately imitate the way the face and mouth move to generate key sounds. At the moment it takes 14 hours, but that could be cut down to just one.
They argue the technology could be used to animate historical audio files and create interactive video interviews with historical icons. Or, it could be used to add visuals to a video call where only low-bandwidth audio is currently possible, such as rural and remote areas with limited internet infrastructure.
If those were the first uses you thought of, however…well you’re a noble and righteous person. While the researchers say that the technology can only make use of words that the person has already said (just not at the place and time shown in the video) it seems to open up some concerning possibilities.
Similar tech has certainly been put to colorful uses.
Other Techniques for Video Duplicity
If you download the video calling app ooVoo right now you can don a digital mask of President Trump and try to convince your friends you’re the most powerful man in the free world (take a look at our ooVoo video chat review to see it in action).
That’s a very crude way of achieving a similar end, but the same basic face mapping has been used to create wearable celebrity masks based on real-world footage. Stanford researchers last year showed how anyone can make President George W. Bush (and Putin and Trump) look foolish during a recorded interview on CNN. The technique has stunningly good-looking results.
Of course, it is not just the face that can be manipulated. There are machine-learning bots independently impersonating people on social media. More worryingly, there’s also a voice-imitating app that can mimic your speech patterns and learn to copy your unique emotional shifts. And we’ve talked before about 3D video conferencing using full body avatars created by stereoscopic cameras.
You could, in theory, team all these tools together and create a wholly fictional version of almost any U.S. president that we have enough audio and video footage of. But the fake president scenario might not be the most likely situation, or the most harmful one. Is a fake president going to dupe you out of your credit card details, or get you fired?
The Video Call Editing Tool Gone Bad
For all that Edward Snowden has told us of the intricacies of digital spying, it’s simple email scams that result in the theft of hundreds of thousands of dollars a year. In part, these are effective because they target an everyday aspect of our lives. We open emails as nonchalantly as we do junk mail.
That’s why it isn’t the fake president but the fake coworker you have to fear once these video editing technologies become commonplace. As the UW researchers have noted themselves, a faked message from the White House can be fact checked by the media and anyone with a search engine, and once we become familiar with the idea we may well grow as skeptical of video as we are of enhanced images online and in magazines.
But what if someone hacked your boss’s account? What if they could watch hours of recorded conference calls kept hidden in cloud storage? What if they used machine learning to recreate this person so accurately they could stage a real-time video conference call in their visage?
What damage could a cyber thief, or even a colleague with a grudge do with a working avatar of your boss at their disposal? Or a digital clone of you?
The duplicate could be authentic enough to pass facial recognition, or voice recognition, barriers and stage entire video meetings in disguise. If you’re a telecommuter who conducts all their business online (like myself) then a world of cloned virtual colleagues is basically your worst nightmare.
A fake president suddenly calling for donations to a new charity may not fool you, but a cloned digital boss asking you to send your account details to a new email address could your empty your online wallet.