Facebook Trained AI To Fool Facial Recognition Systems Even On Live Videos

While Facebook faces scrutiny for its facial recognition practices but the company continuous to research to further develop its artificial intelligence. The purpose of researching in this division is to counter the activities that the company has been accused of.

Facebook AI Research (FAIR) is working on “de-identification” system that is compatible with videos, including live videos. It uses machine learning to change the facial features of the subject in a video in real time. It tricks the facial recognition system by preventing it to recognize the subject.

There are already companies as Israeli AI and privacy firm D-ID, with de-identification technology that works with still images. There are many other facial recognition fooling imagery, known as adversarial examples, that exploits the computer vision software weaknesses to trick it.

Like there is pair of glasses that have adversarial pattern on it which fools the facial recognition system by tricking it to recognize you as Milla Jovovich.

Unlike previous technologies, Facebook trained AI fools the system in real time and on video format, be it pre-captured or live. FAIR claims it is happening for the first time and it is capable of fooling any reputable facial recognition system.

Company also issued a paper in which it explained its approach towards the new technology. It presented a stance that facial recognition can breach privacy or misleading videos can be created using face replacement technology. To control the abuse of the face recognition technology, the company has introduced de-identification methods for videos with a high quality results.


According to reports, Facebook does not plan to use this technology for its commercial products. However, it may be used in future to develop tools to protect privacy of individuals.

Lawmakers and tech companies are already looking for tools that can identify deepfake through software, or regulate the means of spreading fake videos, images and audio.

FAIR is also expanding its research over facial recognition concerns as it could violate human rights and thus needs o be regulated through law enforcement, government or corporations.



Read next: Securing Your Reputation Online With Artificial Intelligence (infographic)

No comments:

Post a Comment