It looks like an ATM machine with a video monitor looming above it.

And it might be the next big technology in lie detection.

Known as AVATAR -- for Automated Virtual Agent for Truth Assessments in Real-Time -- the machine uses an animated face to ask questions, and then employs an array of sensors to determine whether a person is trying to hide something when he answers. The system, developed at the University of Arizona, is still in the trial stages, but early field tests have shown promising results, said Judee Burgoon, a communications professor who has helped develop the technology with funding from the Department of Homeland Security.

“We definitely have demonstrated the proof of concept,” she said, but the machine still needs a lot of refinement, and “we don’t want something adopted that is not delivering what is promised.”

Advertise with us

AVATAR has been used in field tests at a Mexican-American border crossing in Nogales, Ariz., and at an airport and other sites in Europe. In one experiment, security guards in Europe were given false documents and personal stories, and were then examined by AVATAR and by other trained security guards, she said. In that test, AVATAR detected all the imposters; the human questioners missed them all.

But Judith Edersheim, co-founder of the Center for Law, Brain and Behavior at Harvard Medical School, said the problem with such tests is that “telling someone to pretend to be afraid or pretend to be suspicious may not be the same thing as someone who is really afraid or suspicious.”

Still, she said, multi-sensor systems like the AVATAR have the potential of doing a better job than the much-criticized polygraph, which measures breathing, sweating and blood pressure changes. “I’m actually a neuro-optimist,” she said. “Anything that can make us more enlightened and more secure is something I’m in favor of.”

Ms. Burgoon noted that AVATAR is not meant to be lie detector, strictly speaking, but a machine that can detect when someone needs to be questioned further.

“This is not necessarily going to identify whether someone is being truthful, but it could raise flags. A raised [vocal] intonation pattern might lead to a level of insecurity about someone’s honesty, for instance. With gestures, some people who are lying may become more fidgety, and some more rigid. There is this pattern of locking down that many people engage in when they are trying to hide something.”

Advertise with us

Even after all the machine’s signals are taken together and analyzed as a whole, she said, “we primarily see this as a physical support system for the humans who will make the final determination. It is not a truth or lie machine, but instead shows levels of stress.”

In some ways, that makes it a cousin of the polygraph, but at a more sophisticated level. Cameras measure pupil dilation -- a sign of arousal -- as well as facial expressions and gestures, Ms. Burgoon said. Microphones can pick up vocal stress or hesitations, and if a person is asked to type on the machine’s keyboard or use a mouse, it can interpret those gestures as well, all using algorithms that are designed to improve with experience.

It’s based on the idea that the more measurements a device can take and combine, the more likely it is to pick up signs of concealment.

“When you rely on a single kind of measurement, in some circumstances it might be just perfect,” Ms. Burgoon said, “but I think you’re far better off with the checks and balances of multiple sensors. Not only is there no single Pinocchio’s nose, but people differ, and even the same person might show certain signs early on and different signs later. You need to look at multiple signals and piece those together.”

The AVATAR is not the only attempt to outdo the polygraph. Other companies and research projects are looking at vocal stress, facial emotions, and brain signals from functional magnetic resonance imaging or electroencephalograms. The trouble with many of those ventures, said Harvard’s Dr. Edersheim, is that their interpretive computer algorithms are proprietary and can’t be tested by outside researchers.

Advertise with us

This search for a holy grail of lie detection raises one other question: Why aren’t human beings better at detecting liars, especially since evolution has equipped us with extreme sensitivity to other people’s faces?

“Face recognition is not the key” to detecting a lie, Ms. Burgoon said. “It is simply an indication we are attuned to people’s faces, but the person behind the face is doing a lot to regulate the face. Given the fact the face is one of the most controllable parts of the body, that’s how we get misled.”

Mark Roth: mroth@post-gazette.com, 412-263-1130 or on Twitter @markomar