TROUBLE WITH REALITY
We are used to digital fakery, from Photoshopped fashion shoots to blockbuster CGI effects—we know all is not as it appears. The fakery may not be new, but the ease and accuracy with which it can be done using a machine learning system is alarming.
Witness deepfake; this emerged in the fall of 2017, when a number of pornographic videos, apparently featuring celebrities, appeared. All artificial. Deepfake uses a neural network to superimpose faces on to figures in videos. It’s trained using multiple images of the target victim, and can operate in near real time. It was soon followed by FakeApp, a simple application version of which required little programing skill. Now it became possible to easily subvert the verisimilitude of videos shared online.
Despite being banned on a number of social media platforms, deepfake is still widespread. It has been used to create political satire, plain mockery, and put to more devious and hurtful uses, as well as putting the face of Nicolas Cage on famous movie clips for no good reason at all, other than to create an Internet meme.
There are limits—while pretty impressive, close inspection does reveal oddities. But, if you’re scanning through video clips on your phone, a deepfake fake would readily slip by unnoticed. These are dangerous waters, and they make it more important than
ever to always trace the source of something before you give it credence—unfortunately, something that is easy to preach, but not so simple to follow in practice.
There’s more to come: Deep Video Portraits, for example. These go further, and enable you to use the movements and facial expressions of one person using the head of another. Now you can have your fakes move and express themselves anyway you wish, rather than merely being superimposed on to a target video. Full digital avatars of anyone we want made possible from your desktop can’t be that far away. The democratization of Hollywood CGI effects perhaps, but not without ramifications to trust.