New AI Tech Makes Creating ‘Deepfakes’ Possible From Single Image

CYBERSPACE—Researchers working for Samsung in the company’s artificial intelligence labs have created a new algorithm that can animate realistic facial expressions from a single still photo of a human face—such as a social media avatar—and they have now announced their innovation even though they say they are aware it could be used to create “deepfake” videos, according to a paper published by the researchers at Arvix labs.  The term “deepfakes” refers to videos in which a person appears in a video simply by superimposing an image of that individual’s face onto bodies in the video. Using AI algorithms, the creators of a deepfake video can give the face realistic facial expressions and reactions. As so often happens with technological innovations, the first use of “deepfakes” tech occurred in porn, with users creating phony “celebrity” porn videos and posting them in the online discussion forum Reddit, as AVN.com reported.  But the process of animating the celebrity faces to have realistic expressions as might be seen in a porn video was cumbersome and consumed large amounts of computing power as well as the creator’s time. But in a demonstration video posted by the Arvix scientists, the new Samsung AI us employed to create realistic videos of facial movements from the single, still image of the Mona Lisa. “We realize that our technology can have a negative use for the so-called ‘deepfake’ videos,” the scientists write in their paper. “However, it is important to realize that Hollywood has been making fake videos (aka ‘special effects’) for a century, and deep networks with similar capabilities have been available for the past several years. Our work (and quite a few parallel works) will lead to the democratization of the certain special effects technologies.”  The sudden emergence of deepfake videos has raised concern not only that the AI technology can be used to create non-consensual porn videos—superimposing an AI-animated face of a celebrity or any person into an existing porn clip—but also to manufacture fake videos to used for political propaganda and disinformation, as The Verge reported. Donald Trump this week posted a video to his Twitter account of his political adversary, House Speaker Nancy Pelosi, that had been digitally altered to make Pelosi appear to be stammering and slurring her words. Though the video, which was quickly identified as a fake, used standard video editing techniques, AI technology could make such political disinformation videos more difficult to detect—and more convincing to viewers. Photo By YouTube screen capture