A group of researchers from the University of California, San Diego, and the University of Toronto have built a computer that’s better than humans at detecting when someone is faking facial expressions. Trained on enough images of real expressions versus faked ones, the system is able to discern the differences between voluntary and involuntary facial movements that indicate actual pain.
The research is actually not too surprising considering recent advances in deep learning models coming out of places such as Google and Facebook. The more and better training data that these computer vision systems have, the better their models become at detecting the many tiny features that represent any given thing. They’re better than humans in instances like spotting fake expressions because they’re focused on subtleties our brains can overlook.
According to a press release describing the research (the full paper is published in the March issue of Current Biology) the computer vision system was accurate 85 percent of the time compared with a top of 55 percent for human judges. It could have applications in a variety of areas, including “homeland security, psychopathology, job screening, medicine, and law.” Presumably, it could also determine once and for all whether professional wrestlers are acting.