by Amanda McAllister and Navin RamalingamA “deepfake” is an ultrarealistic fake video made with artificial intelligence software. The term is a portmanteau of the concept of machine “deep learning” and the word “fake.” Essentially, it is the end-product of a computer program “learning” the map of the target subject’s face, finding common ground between two faces, and stitching one face over the other in a video editing process.
Manipulating video is not necessarily a recent invention. Hollywood has been doing it many years, such as when film effects were used to make Joseph Gordon-Levitt look like a young Bruce Willis in the film Looper, or the digital recreation of a young Carrie Fisher on actress Ingvild Deila for the Princess Leia cameo in Rogue One. Face-morphing features are also an essential part of multimedia messaging applications like Snapchat.
While the technology may not be inherently illegal or unethical, some manifestations of deepfakes do have the potential to be illegal, to create liability, to spread misinformation, or to violate the privacy of subjects of deepfakes.
For instance, people have been using facial recognition apps and deepfake technology to superimpose faces of well-known celebrities and ordinary people over that of actors in pornographic films or over nude photos to create nonconsensual pornography. The past year has seen several deepfake consumer apps being released permitting its users to create their own deepfakes, including one disturbing app that provided its users the opportunity to create nonconsensual pornography by “undress[ing] photos of women” and making them look “realistically nude.” Fortunately, this app has since been taken down.
