CNET también está disponible en español.

Ir a español

Don't show this again

#BlackoutTuesday Roku Channel launches 100 free live TV channels Facebook employees stage virtual walkout PS5 event delayed Apple stores close amid widespread protests Rick and Morty season finale

Now Skynet can tell when you fake a smile

System developed by MIT researchers can tell the difference between smiles of delight and frustration. It may help in autism therapy.

Video screenshot by Tim Hornyak/CNET

In the future panopticon society of all-seeing robots, don't count on expressing your loyalty to our metal masters with a halfhearted grin.

MIT boffins have already trained computers to recognize real smiles of delight from smiles borne out of frustration. And natch, they can already do it better than us lowly meatsacks.

In a paper published in IEEE Transactions on Affective Computing, Media Lab graduate student Ehsan Hoque and colleagues present some interesting results of experiments related to human psychology and computer vision.

Subjects were filmed while watching a cute baby video or while filling out a long form designed to frustrate them. When they were asked to put on a frustrated expression, 90 percent didn't smile. But when they were filling out the form and suddenly all the data were deleted, 90 percent of them did smile.

One has to admire the Zen-like calm of folks who crack a grin when machines flip them off. I would have cursed at the least. But I digress.

An algorithm used by the researchers tracks the "low-level features" of facial expressions. These can be quantified according to the Facial Action Coding System (FACS), which classifies dozens of facial movements by muscle group.

A Duchenne smile, for instance, involves the involuntary contraction of the zygomatic major muscle, which raises the corners of the mouth, and the orbicularis oculi, which raises the cheeks. It's regarded as genuine, but can of course be faked.

But using the zygomatic major alone, in voluntary fashion, results in the very plastic "Pan-Am" smile, named for flight attendants of yore.

The algorithm was able to detect the frustrated smiles more than 90 percent of the time, while humans performed below chance. In the video below, Hoque says this is likely because the system can focus on the mechanical details of the smile.

The research may have an application in autism therapy. It could be used to help people who have trouble recognizing others' mental states. And, interestingly enough, salespeople and marketers could also benefit.

"Just because a customer is smiling, that doesn't necessarily mean they're satisfied," Hoque said in an MIT release. "The underlying meaning behind the smile is crucial."