Use videos in court to 'teach' computers that detect lies

By using trial videos in high-court trials to "teach" computers, the Michigan team has successfully developed a software that can detect lies with up to 75% accuracy ( human ability to identify lies is only 50%). Based on the gestures and words spoken by the speaker, the software will show whether the person is honest or not without having to touch the body directly like the previous polygraph lie detector.

During the software development process, the University of Michigan team used the Machine Learning technique to "teach" computers a series of 120 video clips of trial scenes and testimonies in dark court trials. High real life. Among them were cases where testimony was exposed, and the cases were wrongly condemned by false testimony.

Computer science professor Rada Mihalcea, who led the study, said: "In the laboratory, it is difficult to create conditions that motivate people to actually lie. Although we can give it to anyone. That money to make them try to lie to someone, change the white instead of black, but that is not enough, so find the real world, in court, where people really have the motivation to lie ".

People who lie often move their hands more when lying, trying to play more clearly.

The videos in the study recorded the testimony of the accused and witnesses, half of them telling the truth, the rest were lies. Sounds including "um, ah, uh" are recorded by the computer, analyzing the frequency of use of the object. At the same time, the machine also considers the body's movement in 9 different areas of the head, eyes, forehead, mouth and hands.

Accordingly, liars often move their hands more when lying, trying to play more clearly. In some cases, the liar also looks at the questioner 's eyes with a frequency more often than the person who is telling the truth, . Based on these lists of behaviors, the computer can detect and count the actual number show, the word gives the conclusion whether the person is telling the truth or not. They said that this machine can do better than people do "we can hardly count how many opposite people are doing something".

The team said that it is continuing to develop a technique to use thermal imaging to deduce the heart rate, breathing rate and body temperature of the person in question, then coordinate with the software to improve the accuracy of the conclusion. without even touching that person. On the other hand, the biggest step of this research is to allow the computer to learn by itself, allowing it to synthesize more data to enrich the list of lie detection features.

This technology promises to be applied in the field of security, defense and investigation work to quickly find the truth. It is unclear whether the technology is widely released, but if it does, women must have had more tools to verify what they have not believed.