A prototype the scientists have built looks at peoples’ words and gestures and doesn’t need to be wired to a person.
The University of Michigan claims that the software was up to 75 percent accurate, using real word data.
The researchers said there are characteristics that show when people are fibbing. They tend to move their hands more, attempt to sound more certain and look their questioners in the eye in an attempt to convince them they’re telling the truth.
The software uses machine learning techniques trained on 120 video clips from coverage of actual trials. The videos include testimony from defendants and witnesses and then compared their results with trial verdicts.