Facebook's facial recognition algorithm is so powerful that it can recognize users in pictures when they're not looking at the camera. Instead, it identifies people based on their hair, clothing style, pose and other physical traits that it compares with old images already stored in its database.
Researchers ran 40,000 Flickr images through Facebook's algorithm and found that the machine was able to identify people with 83 percent accuracy, whether they were smiling into the camera or not. Publication of the final study, which was made public earlier this month at the Computer Vision and Pattern Recognition conference in Boston, coincides with growing tension between privacy groups and the technology industry over the unauthorized collection and storage of Americans' faces.
“There are a lot of cues we use. People have characteristic aspects, even if you look at them from the back,” Yann LeCun, head of artificial intelligence at Facebook, told New Scientist. “For example you can recognize Mark Zuckerberg very easily, because he always wears a gray T-shirt.”
This comes a week after nine prominent privacy groups ended negotiations with the U.S. government and the tech industry after failing to agree on even basic privacy protections around facial recognition. After 18 months of discussions, the sides were still split over whether companies should be able to track people by name in public without their permission.