UB Deceptive Facial Expressions Study

UB Logo

New UB study shows that computers are more successful at
detecting faked emotions than humans are

Dr. Mark Frank
This past year, Dr. Mark Frank , along with researchers from the University of Toronto and the University of California, San Diego (See table below), conducted a joint study to test the the accuracy level of computers and humans detecting faked vs. genuine facial expressions that was deemed one of the University at Buffalo's Most Interesting Discoveries of 2014. The study included 205 human subjects who were shown video clips of people taking "The Cold Pressor Test," or dipping their hands in ice cold water and reacting to the pain (See video below). Some of the reactions were real, and some were not. The participants were then asked to assess whether they thought the reactions of pain in the video clips were genuine or faked based off of the subjects' facial expressions. On the other hand, the researchers conducted the same test on a computer program called the Computer Expression Recognition Toolbox (CERT) that they had developed in order to see whether the computer system or the humans would be more accurate.The researchers found that the computers did significantly better than humans at detecting the deceptive pain expressions.

"Even after training, humans were accurate only 55 percent of the time. However,
the computer was accurate 85 percent of the time" Frank explains.

The researchers found that even after training, the human subjects were only accurate about 55% of the time , putting them close enough to chance probability. On the other hand, they found that the computers ranked at a significantly higher success rate, as they were right 85% of the time . The researchers believe that this is because the machine vision has no human bias, and can instantly detect non-genuine facial signals by quickly picking up on slight behaviors that humans either cannot or oftentimes do not catch. Interestingly, the study found that the most predictive feature of the faked facial expressions was the opening and closing pattern of the individuals' mouths. The fakers' mouths apparently tended to open too regularly and with too much of a pattern, instead of being a sporadic, genuine reaction to pain. The study in it's entirety, entitled "Automatic Decoding of Deceptive Pain Expressions," was published in Current Biology in 2014.

"Often these behaviors are subtle," Frank says. "They're often quick, less than half a second,
less than a quarter of a second in duration, and people often don't see them."

Researchers Involved in the Study:
Dr. Mark Frank
University at Buffalo
Department of Communication
Dr. Marian Bartlett
University of California, San Diego
Institute for Neural Computation
Dr. Kang Lee
University of Toronto
Institute of Child Study