WXXI AM News

UR research on lie detection could help at airports

May 23, 2018

Participants were instructed to lie or tell the truth in order to gather data about the characteristics of human behavior when we lie.
Credit Provided

Researchers at the University of Rochester are developing ways to improve lie detection, with the hope that it could lead to more accurate airport screenings.

Currently, the way a TSA agent decides who is hiding something is pretty basic: Agents have a set of guidelines, called a behavior detection program, to help them seperate truth from fiction.

The guidelines identify behaviors from wearing certain attire to displaying exaggerated emotions to yawning as possible indicators of deceit.

The program has been called into question.

Last year, the ACLU put out a statement saying it leaves too much room for human error, and worse, creates an unacceptable risk of racial and religious profiling.

Ehsan Hoque thinks we can do better. Hoque is an assistant professor of computer science at the University of Rochester. He said there hasn't been a lot of data available to recognize how a person acts when they lie until now.

Ehsan Hoque
Credit Provided

He and his team have been using data science to create a system that detects the characteristics of a lie more accurately. They did this by instructing participants in video chats to lie to each other or tell the truth. Then they recorded the changes in their faces and voices.

"We found that when people are lying," Hoque said by phone, "they are showing 'true smile,' which is when your lip corner pulls up as well as your cheek raises. And these smiles are involuntary, so you cannot really fake it."

Why the smile? One theory is that some people enjoy getting away with a lie.

These expressions areĀ  mostly invisible to humans, but can be detected by computer programs. Hoque said they are still in a basic phase of their research, but he envisions a future in which this technology would be used alongside human screeners to reduce discrimination.