MIT Software Identifies Hidden Frustration Much More Accurately Than Humans

Brad Chacos

The human race's march towards becoming disposable batteries for power-hungry robot overlords continues. A mere day after Intel announced a new project that hopes to bring senses, smarts and adaptable machine learning to computers , word's come in that M.I.T. has developed a program that can detect frustration in human smiles much, much more accurately than people can. That's right, your future robo-master will be able to outsmart you and tell when you're lying.

Via webcam, the test subjects were asked to smile and act frustrated, but canned responses are obviously of limited use, so they also performed tasks designed to evoke true happiness or irritation. The researchers evoked feelings of delight by showing the subjects a happy-go-lucky video, but the way they coaxed out frustration was much more devious; the subjects filled out a long, tedious web form which was rigged to clear when they tried submitting it. Hilarious!

The tests found one other interesting tidbit: a whopping 90 percent of people smiled in frustration when the form cleared, even though only 10 percent of test subjects smiled when asked to mime frustration. Which of the two smiles above do you think is real, and which is frustrated?

The team of researchers developed software that uses small physical clues, such as muscle patterns and smile timing, to determine whether or not smiles are real or hiding frustration. The computer was able to correctly identify frustrated smiles a whopping 92 percent of the time; human observers performed even worse than chance.

The team published a report of their findings in IEEE Transactions on Affective Computing ; has a nice summary of the highlights, while the MIT Media Lab offers more insight into the psychology and science of reading smiles.

(By the way, the smile on the right is the irritated one.)

Around the web

by CPMStar (Sponsored) Free to play