These findings have been presented in a study published by the Facebook AI research team in New York. The study explains how Facebook has been teaching their A.I models to identify and learn images of simple things such as airplanes, animals, and buildings, much in the same way that a child learns to recognize what a car is after having seen one.
In the experiments conducted at Facebook, the firm’s computers were able to auto-generate (this is, “draw”) 64×64 pixel images close enough to the real thing that they persuaded — or fooled — human eyes nearly half of the time.
And image generation isn’t the only thing Facebook is cooking up in their A.I. lab. Facial recognition is also part of the future of A.I., and it’s getting eerily powerful. Controversy has always surrounded this capability, as many believe that storing your face in a memory bank is an invasion of privacy if done without consent.
Another study published by Facebook describes the increased ability of their A.I. computers to recognize faces based on things like hair style, clothing, and body shape. The new technology is accurate about 83% of the time, which is leagues ahead of current standards that are not only less accurate, but require clear full frontal views of faces to identify names.
You’ll see this in action whenever you upload new photos to Facebook when your friends’ faces are automatically identified, but only in the clearest of pictures. This new technology is capable of recognizing partially obscured faces, and under some circumstances can even identify people from the back of their heads.
These capabilities are far from ready to be fully implemented, and it may be years before we see this sort of technology being used on our Facebook pages, but the company is working on it.
