Facebook AI can 'hide' people from facial recognition
It could help fight deepfakes and other privacy issues.
Facebook has already stopped using facial recognition by default, but now it might have a way for people to dodge facial recognition altogether. Its researchers have developed an AI system that can "de-identify" people in real time, including live videos. The approach pairs an adversarial auto-encoder with a trained face classifier to ever-so-slightly distort a person's face in a way that confuses facial recognition systems while preserving a natural look that stays recognizable by honest-to-goodness humans. You might see tiny differences in Jennifer Lawrence's face, for instance, but you won't have any doubt that it's her.
The AI doesn't need to be retrained for different people or videos, and produces only a "little" time distortion.
Don't expect this to reach Facebook in the near future. A spokesperson told VentureBeat that there were no plans to implement the research in its products. With that said, some of the practical applications are already clear. This could be used to automatically thwart third parties using facial recognition software to track people's activity or generate deepfakes. Theoretically, you could upload a video of a house party without worrying that your friends' facial data will be harvested against their will.
