Think you’re a good liar? Well, soon, the jig might be up: Researchers have developed new technology that reads subtle facial expressions to sniff out bullshit better than humans can.

Machines that can read human facial expressions aren’t new. But machines are getting a lot better at reading human microexpressions. “Microexpressions” are what psychologists call extremely slight, fleeting facial movements that flicker across someone’s visage, like when they’re lying through their teeth. Now, researchers at Finland’s University of Oulu say they’ve produced and tested the first AI system that’s better at spotting microexpressions than humans are. The paper was submitted earlier this month, released on arXiv, and is awaiting publication.

Advertisement

“Microexpressions tend to occur when individuals hide their feelings under conditions of relatively high stakes,” MIT Technology Review reports. To make AI that can suss out those expressions, the researchers assembled a database of footage of 20 people watching an emotionally charged video, who were forbidden to show any emotions on their faces. Using a powerful camera that captured images at 100 frames a second, the team scrutinized subjects’ faces for the exact moment that even tiny changes appeared.

The researchers snapped 164 of those valuable, rapid, involuntary, less obvious microexprssions from the group’s faces, and then matched each one to the content from the video that elicited it. Voila: A machine algorithm that can spot microexpressions similar to the ones the team teased out from the experiment. Later, 15 humans were shown images of the captured microexpressions, and were then asked to ID them in raw video of the subjects’ faces. The machine significantly bested them.

Advertisement

One does have to be a bit skeptical, though. While emotion-reading robots are all the rage in robotics right now, can a machine really do a better job sizing up human emotions than another human could? It seems farfetched, but the algorithm’s precise detection capabilities might come in handy for psychotherapists and law enforcement. And maybe one day, it could help you at the poker table, too.

Advertisement

[Cornell University Library via MIT Technology Review]

Top image via Shutterstock