From the polygraph machine to purported “truth serum,” law enforcement and national security agencies have spent significant effort developing and researching tech solutions for determining whether or not a suspect is telling the truth. A recent study from the University of Padova claims to be able to do just that, all by tracking the movements of a computer mouse.
Scientists and researchers at the Italian university performed the study as a smaller piece of a larger effort to develop artificial intelligence. The lie detection system began by having subjects respond to questions, both truthfully and dishonestly, via clicking the answer with a computer mouse. The results were then fed to the artificial intelligence algorithm, which had been trained to make decisions upon submitted data.
The study found that respondents who were instructed to answer questions truthfully clicked the accurate answer directly and tended to follow a straight path with the mouse. The lying respondents, however, tended to veer, hesitate, or take a less direct path before clicking the dishonest answer. Labeled examples of each were provided to the a.i. algorithm, which then learned to associate the mouse-click behavior with truthful and dishonest answers.
Giuseppe Sartori, professor at Padova and the lead researcher on the project, told the press that the betraying behavior ultimately boils down to how the human brain works.”Our brain is built to respond truthfully. When we lie, we usually suppress the first response and substitute it with a faked response,” he said.
Sartori has high hopes for the his lie-detecting method. He envisions the project being used to catch pedophiles and online sexual predators, insurance fraudsters, and even terrorists. The team concedes that the algorithm isn’t foolproof. A coached or otherwise well-prepared individual could be trained to respond quickly with deliberate lies, but the team also suggested that liars would likely slip up when answering offhand or unrelated questions, such as submitting their astrological sign.
The artificial intelligence algorithm is entirely dependent upon the sample size available for training. Sartori and his team intend to build the data set to ensure the validity of their results. The team also has plans to expand the project, beginning with determining if lying affects how individuals type in responses via keyboard.