A View from Christopher Mims
Software Better at Detecting Frustration Than Humans
And it’s not because subjects have been trained to “smize”.
Did you know that you smile when you’re frustrated? Starting at 1:20 in the video below, witness a behavior that you may find novel – and doubly so because you’re a human being who is exquisitely tuned to reading the emotional expressions of others.
By training a feature-detecting algorithm to decompose subjects’ facial expressions into individual “action units,” M. Ehsan Hoque of the MIT Media Lab discovered not only that smiling is quite common when test subjects are frustrated, but also that software is better than humans at differentiating frustration smiles from happiness smiles.
“For frustrated smiles, humans performed below chance, whereas the algorithm did better than 90 percent,” says Hoque.
Subjects’ frustration smiles included activation of the groups of facial muscles known as action unit 12, or raising of the corners of the lips, as well as action unit 6, in which cheeks are raised.
No word on whether or not the smile detector can differentiate between real smiles and Tyra Banks’ patented “Smize.”
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today