Making sure if a person is lying or not can be pretty difficult. It relies on a multitude of factors that we, as humans, as well as polygraph machines, can’t really detect very well. This is what scientists at the University of Michigan are trying to solve, by creating lie-detection software based on real-life cases.
The team behind the project
The team of researchers from the University of Michigan – Flint, led by Dr. Rada Mihalcea, professor of computer science and engineering, and Dr. Mihai Burzo, assistant professor of mechanical engineering, is trying to create a software that is able to detect lies better than humans or any other lie detection device invented up until this point.
And they’re doing a pretty good job of it so far, managing to score positive results in about 75% of the cases, unlike humans, who usually get it right in about 50%.
Yes, lie detection is that much of a risk. With us being right in only 50% of cases, it’s a good thing that we’re not using human lie detectors in court cases, otherwise a lot more innocent people would be in jail.
How the software works
Using footage from actual cases, the team of scientists is programming the software to recognize human lying patterns.
They are using parameters like motions of the head, face, and hands, looking directly at the questioner, the use of vocal fill such as “um, ah, and uh”, and even heart rate, respiration rate and body temperature fluctuations.
All of the data is collected remotely, via thermal imaging and facial recognition, allowing for a far less invasive procedure than using a polygraph.
The researchers used a series of 120 video clips from actual trials, in collaboration with the The Innocence Project, an organization meant to help speed up the release of innocents that were mistakenly convicted.
Using machine-learning software, like Google’s DeepMind, the scientists managed to instill in the machine a more powerful sense of truth than in most humans. And this shouldn’t come as a surprise, since humans are known to let their emotions get in the way.
The use of footage from actual trials is important, because this allows the machine to register subtle movements that would have most likely been absent in a laboratory experiment, as the subject wouldn’t have been as motivated to lie.
The project is still in development, but the researchers behind it hope to have it ready in a relatively short matter of time.
Image source: Wikimedia