In a country where real is fake and fake is real, how can you know what is true anymore? There’s a fake president (allegedly hacked into office by the Russians) and 'Fake News' and a fake White House (that relays Fake News, only this time from the President). And if that doesn’t go fake enough, you also have 'deepfake', where videos purporting to be someone are actually not that someone at all. So how can you know what to believe?
In April 2018, BuzzFeed published a video of former US President Barack Obama cursing and calling President Donald Trump names. Thing is it wasn't Obama at all. It was the voice of director and actor Jordan Peele, which had been inserted into an original clip of Obama, effectively creating a 'deepfake' - a video of someone saying or doing something that didn't happen.
Two days ago, a Belgian socialist party circulated a ‘deep fake’ Donald Trump video in which the U.S. President apparently calls on the country to follow America’s lead and exit the Paris climate agreement. Toward the end of the video, Trump says, “We all know climate change is fake, just like this video."
Readily available video-merging program from websites like FakeApp scan videos and still photos of one person and paint that person’s features onto another person in a separate video. Using artificial intelligence technology, the programs can replace faces down to the movements of eyes, mouths, and heads.
It’s an evolution of the way Adobe Photoshop, created 30 years ago, to alter still images. In fact, one popular online pastime predating deepfakes is a series of memes and GIFs depicting actor Nicolas Cage’s face photoshopped into everything from Harry Potter to Michelangelo’s 'The Creation of Adam.' DeepFake videos go a step further and transmogrified Cage onto Lois Lane, Luke Skywalker, John Kelly or yourself. And nowadays this video-merging technology is accessible to users who would have one time lacked the money to skills to buy and use it.
Banning deepfakes can violate the First Amendment, but they can cause some very real trouble. Imagine if a video of Donald Trump goes rogue with the U.S. President threatening to unleash nuclear war on an unwitting world.
This returns us to our original question: how to tell false news, or false videos, from true.
How Technicians Recognize DeepFake
Here’s what Siwei Lyu told us: “One clue we use,” said the digital forensic scientist, “is an algorithm that identifies tiny fluctuations in the skin colors from changes from the blood flow. As you know, every person has a heartbeat, so this heartbeat causes a tiny change of skin color.”
“Now with a technique that is developed by researchers from MIT, we actually can enhance the video after the fact and use that as a signal to differentiate…If you have a video of a real person, you would see this signal prominently. On the other hand, on a generated fake video, the signal of this kind of physiological phenomena will be much weaker.”
The challenge is that once you have a system that can detect a fake, you can train your system that creates fakes to counter that system. So it’s a cat and mouse game, where robber learns from cop to beat cop.
Advice from DeepFake Society
Better advice ironically came from the director of the DeepFake Society that curates the best of those videos and has more than one million views since it launched in February. That site doesn’t allow pornography but has videos like one showing Trump and North Korean leader Kim Jong Un as each other.
When SFChronicle reached through the contact information box on the site, a man called back declined to give his name because he feared the stigma surrounding deepfake pornography could jeopardize his web programming job. “The malicious implications of deepfake,” he revealed, “are absolutely terrifying. You can put any politician doing anything anywhere. Even if it is fake and it gets out, it’s going to ruin somebody. Most people don’t see a report and go out and do their own research. They just take it at face value.”
And so, ultimately, our best tip points to one thing: Do your own research.
Learning From Fact Checkers
How do fact checkers do their job? They isolate a claim with something that can be objectively verified. The video of Obama bad mouthing Trump falls into that realm. They seek the best primary sources in that topic, namely they look for objective news reports that support the video. The reports span the political platform and are non-partisan. Last, these factcheckers assess whether their sources match the video or refute it. That's the framework for fact checkers, and that’s the method we can use for ourselves.
In an upside down, topsy-turvy, world which recalls French philosopher Descartes’ suspicion that demons warp our sense of reality, we cannot be careful enough. We’re inhabiting a sci-fi reality and its stranger than you’ve ever imagined. They say seeing is believing, but that old proverb is outdated.