The computer was trained to recognize deception in court
American developers have created a system of artificial intelligence of DARE, which can recognize the lies of the defendant by his face and voice.
Testing showed that the program copes with this task better than humans — the computer has guessed the lie to 90 percent of cases. The program will be presented at the conference on artificial intelligence AAAI-18, a description of how it works can be found in the paper on ArXiv.org.
People in everyday life is bad enough to recognize the deception. Studies show that on average, only 54 cases of untrained subjects can correctly identify whether honest with them other participants of the experiment. However, the lie produce different characteristics is not only the change in physiological indicators of the body (heart palpitations or sweating), but also external characteristics.
For example, even small changes in facial expression can reflect the emotions that the person wants to hide.
Modern machine learning systems have great potential for detecting deception. This is largely due to the fact that the computer is trained on a large dataset, and it allows him to notice small details that people just don’t pay attention. So researchers at the University of Maryland and Dartmouth College have created a special algorithm-Poligraf DARE.
On the basis of the DARE lies a number of existing and pre-trained classifiers and models. For the training algorithm was also used the database containing 121 a video mock trial. The system to track changes in facial expression, voice and speech. The principle of the program is shown in the picture below.