Suffice to say, traditional methods such as the polygraph are prone to high levels of errors, but researchers believe a new automated approach can prove considerably more effective.
The system has been trained to use video footage from real court cases and considers things such as the words used and gestures made during speeches. When the system was put through its paces, it proved accurate roughly 75% of the time, thus providing a more accurate means of detecting lies than human observers.
Looking for tell tale signs
The software was able to detect a number of clear signs that someone is lying. For instance, they appear to move their hands more frequently and attempt to sound more certain. Interestingly, they also attempted to make eye contact more often than usual.
How we deceive
The researchers trained the system using 120 clips from actual courtroom trials via The Innocence Project, who attempt to exonerate the wrongfully convicted.
The team believe that this real world aspect of the work makes their project stand out from other attempts to decipher deceit.
“In laboratory experiments, it’s difficult to create a setting that motivates people to truly lie. The stakes are not high enough,” the authors say.
“We can offer a reward if people can lie well—pay them to convince another person that something false is true. But in the real world there is true motivation to deceive.”
The researchers transcribed the footage and then mined the data for usage of particular words, or even categories of words. In addition to this, they analyzed the gestures used via a coding scheme for interpersonal interactions that scores nine different motions of the eyes, head, mouth, and hands.
“People are poor lie detectors,” the authors say. “This isn’t the kind of task we’re naturally good at. There are clues that humans give naturally when they are being deceptive, but we’re not paying close enough attention to pick them up. We’re not counting how many times a person says ‘I’ or looks up. We’re focusing on a higher level of communication.”
What give liars away
The analysis found a number of behaviors that were common in liars. For instance, they were three times more likely to scowl or grimace than those telling the truth.
They were also nearly twice as likely to gesture with both hands than those who were being honest and were more likely to fill sentences with ‘ums’ and ‘ahs’.
This will eventually be paired up with various physiological measures, such as heart-rate, body temperature and respiration rate, all of which can be gathered using thermal imaging technology.
Whilst in this initial phase, the researchers themselves classified the gestures, they hope in time that the computer can do this automatically.
It’s certainly an interesting project, and better lie detection would certainly be a valuable tool in all manner of scenarios. It will be one to watch with interest.