This software knows if you lie – Sciences et Avenir

DOUBT. Imagine you are interrogated. You do not perhaps realize, but you tend to look your interviewer in the eye right. You also shake both hands simultaneously. And you pretty nod. Not to mention the grimaces, occasionally, disfigure your face. Know this, if all this, you are not guilty of something, you’ll at least suspect. And not need a lie detector to the former connected to your pulse. It took was a camera and software to give rise to doubt

Elocution, gestures and facial movements are analyzed

Four researchers from the University of Michigan, the United States, have indeed developed a prototype supposed to allow, over time, to determine whether someone is lying or telling the truth, simply scrutinizing his speech, his gestures and facial movements. The program is under the technique of machine learning , that it learns by analyzing a set of data and improves its efficiency As he analyzes new data.

BAD LIARS. In this case, the software was trained with 120 authentic videos filmed trial verdict with known (and which are not judicial errors!) and for which we know who lied and who told the truth during the hearing. The advantage of the process? Avoid having to recreate a false position in laboratory trials where a speaker would pretend to lie. Researchers estimate that in this type of context, “liars” play too bad lie because it lacks the inherent motivation at trial: save himself

The videos used show both suspects and! witnesses and, in half the selected cases, people were lying. Researchers fully transcribed hearings by including hesitations and onomatopoeia (the “uh” and “bah”), to determine how to express a liar, what are his mannerisms. Ditto for the gesture, divided into nine categories of head movements, eye, mouth, hands and eyebrows, each corresponding to a level lie.

These are all parameters that have been integrated software. It shows that the speaker set is a sign of probable lie (70% liars did, against 60% who were telling the truth). Shake hands well (40% against 25%), as if twisting the facial features (30% against 10%) (see details of the criteria here).

THERMAL IMAGING. The first results are promising as the team claims to have achieved efficiency rates of 75% device. Research is however not complete because should soon include other parameters such as heart rate or that of respiration, there also data remotely sensed by thermal imaging.