A team of researchers at Tel Aviv University is pioneering a new lie detection technique that can theoretically use facial recognition to determine whether or not someone is telling the truth. The current version of the system uses wearable electrodes to record micromovements in people’s facial muscles, and the researchers are eventually hoping to link those movements to observable facial expressions that could be captured with a facial recognition camera.
As it stands, the movements captured by the surface electromyography (sEMG) system are usually invisible to the human eye. With that in mind, the Tel Aviv system is already outperforming human subjects in initial experiments. The system spotted liars with 73 percent accuracy, while humans knew when someone was lying anywhere from 22 percent to 73 percent of the time, depending on the individual.
While those results are promising, the researchers have only looked at very simple, binary scenarios, and stressed that their solution is not yet ready for real-world situations. Their first experiment involved 40 volunteers, who sat across from one another and either repeated a word piped in through a pair of headphones, or came up with a new word to trick their partner.
Future experiments would look at more complex scenarios. For example, participants might be asked to tell longer stories, or to engage in other forms of deception, such as lying by omission or through equivocation. The researchers are hoping that the system could replace polygraph tests in the criminal justice system with enough refinement.
Until then, the researchers found that some people twitch their eyebrow muscles when they lie, while others moved the muscles in their cheeks. However, they also learned that people who are good at fooling humans were just as good at fooling the machine, and that practiced liars were often able to best the electrodes and the facial recognition system.
The facial movements the researchers are looking at are involuntary, and typically last for only 40 to 60 milliseconds. Over time, the researchers want to use machine learning to get rid of the electrodes, and develop a system that relies only on facial recognition.
Of course, any such system would need to be tested extensively to mitigate concerns of bias. The fact that good liars didn’t get caught is also concerning, insofar as it suggests that it may be functionally impossible to objectively determine whether or not someone is being deceitful with 100 percent accuracy.
The Tel Aviv study nevertheless reflects a broader interest in using facial recognition to find out more about the mental and physical state of subjects. Most notably, researchers at Okayama University recently used facial recognition to study the effects of Parkinson’s disease.
November 24, 2021 – by Eric Weiss