Facebook has had a stormy relationship with facial recognition, currently being locked in a multi-billion dollar lawsuit over its facial recognition practices. But this hasn’t prevented the social media giant from it innovating with AI, and showing how machine learning can help hide you from facial recognition.
The technology has been developed by Facebook AI Research, And is described as a de-identification system. The innovation can be applied to all video, even live video in real-time, and it works by tinkering with key facial features in a video subject. This enables machine learning to fool a facial recognition system, making identification of the subject impossible.
This is just one example of the de-identification technique, with many companies also promoting similar solutions. For example, the Israeli AI and privacy firm D-ID has already dedicated itself to providing de-identification technology for still images.
Another example of technology that can fool facial recognition systems has been dubbed ‘adversarial examples’. This innovation works by exploiting weaknesses in the way that computer software identifies facial characteristics. This means that so-called adversarial patterns can be printed on to, for example, sunglasses, effectively completely deceiving facial recognition systems.
While these previous attempts to circumvent facial recognition were technically impressive, Facebook takes the practice one step further by reportedly achieving its deceit in real time. Facebook AI Research claims that this is a pioneering form of technology, and the first time that this has been successfully achieved in the industry.
Facebook has demonstrated the technique in a video featuring the actress Jennifer Lawrence, and although the social media company has been criticized for its own practices, Facebook envisages that the technology will eventually offer privacy advantages. “Face recognition can lead to loss of privacy and face replacement technology may be misused to create misleading videos,” a paper authored by Facebook, and cited by VentureBeat, notes.
It seems that Facebook has no immediate plans to utilize any of the facial recognition technology in any commercial products. But the research could influence tools that are developed to protect privacy going forward. The technology involved could also inform the artificial intelligence industry, which continues to work on ways to combat the spread of fakery, and the sophisticated tools that are used to create them.
‘Deepfakes’, as they are referred to by law enforcement, can help various individuals spread fake videos, images, and audio. This is currently very difficult to address, but it is hoped that the techniques used by Facebook in creating this facial recognition fooling system can be used in the fight against crimes in this field.
A US federal court recently overruled Facebook’s attempts to stymy a class action lawsuit worth $35 billion, which alleges that the social media corporation has illegally collected and stored biometric data, without acquiring sufficient consent.