Fake news is a serious problem. At best, it misinforms — at worst, it can undermine democracy. And if text-based fake news isn’t bad enough, a team of computer scientists have published a new paper demonstrating how they can fake video using artificial intelligence.
The team of researchers from the University of Washington have detailed a method for faking video in a recently published piece titled “Synthesizing Obama: Learning Lip Sync from Audio.” Essentially, the team was able to manipulate audio and video of President Obama to make it look like he was saying a sentence in a completely different time and place than it was actually stated — you can see the results below.
Instead of copying and pasting Obama’s body in a different location, the researchers used a AI neural network to analyze dozens of hours of video footage to study how Obama’s mouth moves when he speaks. Their computer system could then digitally manipulate the former president’s mouth to “say” any word or phrase that they wanted him to. The team demonstrated the capability with rather innocuous examples — they didn’t make things up for President Obama to say, but they did “cut and paste” sentences out of separate speeches, and were able to change the order of the words he was speaking.
The results are, obviously, extremely worrying. While the researchers only used words that Obama himself said, similar methods can be used to create video of important individuals saying basically anything. In a worst case scenario, it’ll only be a matter of time before we see a slew of faked videos of politicians, journalists and other authority figures that aim to mislead and misinform. We already know how damaging fake news can be, so the best course of action is to be prepared for it. Be skeptical, only trust information from verified outlets, and please, double-check the authenticity of something before you share it.