Lie Theory: Artificial Intelligence Can Do It!
Researchers from the University of Maryland and Dartmouth College has trained artificial intelligence to recognize lie. The artificial neural network has a name DARE (Deception Analysis and Reasoning Engine) and it copes with lie detection by reading mimics and recognition of voice timbre. Only five minor changes in the expression of people’s faces are believed to indicate that someone is lying. They are frowning, raising eyebrows, twisting the lip, protruding the lip and turning the head.
The program evaluated the veracity of the testimony on 15 video recordings from the court. During the tests, the program correctly identified the lie in the answers to questions of the judge in 92% of cases. For people, this index was equal to only 81%.
A database containing 121 videos with a staged trial was used for the training of the algorithm. However, the authors of the program point out one limitation: systems rely on staged trials. In real life, defendants can lie much better, because they are more motivated.