Facebook AI alters faces in videos to thwart face recognition AI

Facebook, along with Amazon and, of course, Google, is facing no small amount of opposition and inquiry into its use of face recognition systems. Primarily applied for things like tagging family and friends in photos, the social media giant is often accused of misusing that technology for less savory uses. Now the company's own researchers are developing AI to combat other AI, perhaps including Facebook's, that try to identify and tag faces in videos.

While the Internet seems to be so fascinated by the entertainment value of some "deepfake" apps and services, others are already raising alarms over their privacy and accuracy implications. It might now be too easy to spread misinformation and is almost extremely trivial to identify people just from photos. Especially the ones they willingly submit only to swap with some other face.

This has given rise to a new industry that developments AI to fight AI, specifically face recognition AI. But while almost all of those have focused and gained success in still photos, Facebook AI Research's new system is perhaps the only on that can do the same for videos, even ones happening live.

This AI uses an adversarial autoencoder to apply slight distortions to a person's moving face. Although the faces are still recognizable to our human brains, face recognition software can be thrown off by those changes. Best of all, since the AI doesn't have to be retrained for each video, it can, in theory, be applied to any live video feed.

Facebook has made no indication of where and how it will use the technologies produced by this research. Other de-identification companies will probably try to develop their own alternatives, in case Facebook decides to again use this new AI for its own gain and profit.