Facebook's facial recognition algorithm is so accurate that it can correctly identify people who have their back turned to the camera.
By looking at their clothes, hairstyle, body shape and pose, Facebook can work out which of your friends is in any photograph you show it. The experimental algorithm isn't available to the public yet, but with a claimed accuracy of 83% it probably isn't that far away.
Yann LeCun, head of artificial intelligence at Facebook, wanted to create a system which can identify people when their faces are hidden – something humans are very good at, because we use other features like height, build, hair and clothing to help recognise our friends.
Speaking to New Scientist, LeCun said: "There are a lot of cues we use. People have characteristic aspects, even if you look at them from the back. For example, you can recognise Mark Zuckerberg very easily, because he always wears a grey T-shirt."
The research team downloaded 40,000 photographs from Flickr and ran them through a complex neural network; some photos clearly showed the subject's face, while others were obscured entirely. Facebook's experimental algorithm accurately recognised people in these photos 83% of the time.
The race to create the most accurate recognition algorithms also includes Google, which recently updated its Photos application to include a feature which categorises photos based on their content, which received widespread praise from users and the technology press alike.
But such powerful software is not universally admired. The ability to recognise people who are perhaps deliberately hiding their face from the camera is a potentially controversial one. Ralph Gross, a privacy protection researcher at Carnegie Mellon University, said Facebook's software "will certainly concern people...now is a time when it's important to discuss these [privacy] questions".