University Times

Could terrorists be thwarted by the automated equivalent of a nosy neighbor who peeps through the curtains at what’s happening on the street outside?

Kevin Murphy, chair of Penn State’s Department of Psychology and director of the school’s Center for the Behavioral and Social Science of Terrorism, thinks so.

In a March 16 talk sponsored by Pitt’s Center for National Preparedness, Murphy spoke on strategies for automatically detecting anomalous behavior in much the same way an overly interested neighbor takes note when something unusual is going on nearby.

Everyone’s familiar with such neighbors, but learning how those people do what they do and how the process might be copied using cameras, sensors and computers is a problem Murphy is studying in an effort to improve security.

Nosy neighbors have no difficulties noticing when something is different in their corner of the world, but how might an automated system mimic that expertise in sensing when something simply doesn’t belong?

“What should we be looking for and how should we be thinking about this data?” Murphy asked.

The problems are tricky. How does a system detect and determine what sort of behavior is expected versus what’s suspicious?

“People do this quite easily,” Murphy said. For nosy neighbors, the process is something like “I’ve never seen this person. I’ve never seen something like this in my neighborhood.” Translating those ideas to an automated system that can define what is normal in an area under surveillance and develop measures for determining when an anomaly has occurred is not as simple or as immediate as the gut reaction of a live human being familiar with a particular area.

Coming up with the proper tests is important. The biggest potential problem, Murphy said, is “We may spend 10 years developing a project that does a great job of doing the wrong thing.”

Categorizing behaviors quickly and matching them to a list of what’s normal in a particular circumstance may be one way to ponder the problem.

Determining individuals who don’t fit in, that is, identifying and categorizing features of people who don’t fit in a certain situation and therefore merit a closer look may be another.

Or differentiating targeted behaviors from a list of other possible behaviors that can be identified may be a way of approaching the problem.

Among the possibilities being researched are detection strategies that analyze how people walk. Gait analysis can detect whether a person is carrying more weight than normal, perhaps from a hidden explosive belt. Kinesiologists are working on ways to translate what digital cameras see into a model of rhythmic movement to detect when someone’s walk is out of the ordinary.

“Am I walking around as if I’m carrying 40 extra pounds under my jacket? If the answer is yes, someone had better go ask me to open my jacket,” Murphy said.

Other avenues to consider would be developing ways of detecting furtive behavior in order to be able to single out someone who is trying to evade detection, Murphy said.

If a person is trying to sneak through Customs, for instance, he or she may walk or talk differently, particularly if made aware that authorities are watching. “If you try to act natural, you overdo a lot of things,” he said.

Researchers are looking into detecting people’s reactions to relevant stimuli and images. For instance, displaying images of improvised explosive devices (IEDs) or puffing trace scents of bomb-making materials could trigger involuntary reactions in people familiar with those devices, perhaps leading them to be singled out for closer scrutiny in an airport security line. That strategy would require a system to identify reliable behavior indicators that would show such stimuli are familiar to the individual, Murphy said.

Murphy said that creating situations that would nudge so-called ‘bad guys’ into involuntarily reacting and tipping their own hand may have merit.

For instance, “You literally never forget smells,” he said, explaining that if a stimulus such as a minute hint of the scent of bomb ingredients is put into the air, the physical reaction that the recognition of the smell would trigger in a bomb-maker could be measured, while the subtle scent would go unnoticed by those unfamiliar with it. “You’d never detect it unless it’s familiar,” he said. Ideally, the test would expose someone who may be on his way home from making a bomb, but not those with no exposure to such materials.

Reactions might include changes in facial temperature, heart rate, blood pressure or changes in gait or skin conductivity, all of which are possible to measure remotely.

The “guilty knowledge paradigm” used in polygraph testing is related. The tester has a particular idea of what is being looked for and creates a situation that increases the likelihood of creating a reaction.

“The biggest challenge is what to show,” Murphy said. In an airport, perhaps an auditory stimulus such as an announcement would be useful. “‘Would all terrorists report to Gate 2-A?’ They probably won’t go, but you will get a reaction,” he said.

All the approaches have merits as well as problems and it seems as though there are more questions than answers.

Researchers need to determine what’s a relevant stimulus. Should you show one stimulus multiple times or several different ones? How many stimuli are needed to get the correct balance of sensitivity and specificity — not taking closer looks at too many people while not letting possible terrorists slip through? What sensors are best? What cultural factors are or are not relevant?

In short, “What’s the right stimulus to pin down the right people? We don’t know the answer to that,” Murphy said.

“We’re just starting to map out the right questions and we don’t have any answers,” he said. “Long term, looking at the end of this project, we’ll probably have an idea of the questions to be answered. Getting a machine to do what a nosy neighbor would do is way further down the road.”

—Kimberly K. Barlow