A Startup Wants to Use Eye Tracking to Detect If Syrian Refugees Are Terrorists
Converus has been pitching its EyeDetect software in Washington as a potential solution to fears about Syrian and Iraqi emigres.
Have you committed any acts of terrorism?
Do you have ties to ISIS or al Qaeda or any other terrorist organizations?
Is your intention for entering the country to commit an act of terrorism?
These are questions a Utah startup hopes to ask Syrian and Iraqi refugees after hooking them up to EyeDetect, a new kind of lie detector that measures eye movements in order to detect deception.
The company, Converus, is using the Syrian refugee crisis as an opportunity to pitch EyeDetect as more accurate and easier to use than the polygraph. EyeDetect is also more adaptable to mass screening than polygraph tests, the company says, making it ideal for vetting the 10,000 emigres expected from the war-torn region.
Converus representatives have been in Washington demonstrating the system to FBI, NSA, Secret Service and members of Congress, according to the company, and pitching it as an optimal way to screen visa applicants and refugees.
How can we catch evildoers without inculpating innocent people?
"Everyone is talking about the need to vet refugees, but no one has a viable solution—especially for effectively screening large groups of people who don't have any personal identification records," Converus President and CEO Todd Mickelsen said in a statement. "EyeDetect has been tested, it works, and it's available to implement immediately."
Converus says EyeDetect is 85 percent accurate, and could reach 98 percent accuracy when combined with other methods. The company says it is already being used in Latin America and in ten US states to vet employees in "trust" positions within government or law enforcement. Might this be the future of lie detection?
Polygraphs remain our best tool to detect deception, somehow. Despite nearly a century of controversy, it's the only gadget or method deemed credible at a federal level. When used by a skilled examiner, with questions tailored to a specific incident, it can differentiate a lie from the truth with accuracy far greater than chance, according to the National Academy of Science. But when it comes to screening large numbers for general "credibility," the polygraph is unsuitable, supported by "scanty and scientifically weak" evidence, according to the same study.
The ACLU has long lobbied against the use of polygraphs in court and in employee screening, due to its inherent invasiveness and issues with accuracy. It's also been criticized for being too easily fooled by savvy liars. Lie detector use by private employers was banned in 1988 for those reasons, and in 1998 the US Supreme Court ruled that judges are not required to admit polygraph evidence in court. Excluding polygraph evidence ''is a rational and proportional means of advancing the legitimate interest in barring unreliable evidence,'' Justice Clarence Thomas wrote in the majority opinion. As of 2009, 30 states did not allow polygraph results to be admitted in court, according to a review by a New York-based polygraph service.
Efforts to improve on lie detection with new technology such as brain scans have therefore been greeted with some skepticism. The sudden demand for mass screening of refugees, however, may open the door for EyeDetect.
Back in December, a California Congressman introduced a bill meant to allay the fears of Americans who are worried that accepting refugees from Iraq and Syria could potentially open the country to terrorists. Among other provisions, the Secure Accountability for Emigres and Refugees (SAFER) Act, H.R. 4291, would require all Syrian and Iraqi refugees to pass a lie detector test. It's unclear if the bill will gain traction, but the fear of admitting malicious actors is real—and Americans may be much more willing to overlook concerns over privacy and accuracy when it's a question of protecting the homeland.
In 2002, John Kircher, an esteemed psychophysiologist and co-inventor of the computer polygraph, went on a hiking trip with a few friends and colleagues, some of whom were cognitive and educational psychologists. Out of their conversation came the question of whether eye-tracking cameras, used most commonly to study reading comprehension, might be helpful in detecting deception. This question would lead to over 12 years of research, and, eventually, the EyeDetect.
"For many years we've known the pupil is diagnostic," Kircher told Motherboard. "Under cognitive stress an ensemble of physiological changes takes place, including a dilation in pupil size." Since telling a lie is more cognitively demanding than telling a truth, they posited that the liar would be betrayed by changes in their eyes.
After returning from the hiking trip, Kircher formed a team with two cognitive psychologists, Dan Woltz and Ann Cook, all from the University of Utah. In 2009, they tested the eye-tracking cameras in a mock crime experiment with 40 participants. Half of the test group was randomly assigned guilty, and told to go perform one of two crimes—filching $20 from a secretary's purse, or stealing credit card info from a student's computer. The other half of the group was assigned innocent. Everyone was told they were suspects, and was promised a financial reward if the system deemed their stories credible. They then read and responded to true or false questions about the crime. The researchers found that the eye-tracking cameras accurately assessed who was lying and who was telling the truth around 80 percent percent of the time. Subsequent tests by Kircher largely confirmed those results.
Simply thinking about a difficult math problem makes your pupils dilate.
"We were predicting that cognitive load from lying would dilate pupils," says Kircher. "And we found that. It's more difficult to tell a lie than tell the truth. It surprised us, actually. Over your career you have all these ideas that you think are fantastic. Very rarely do they pan out. This is one case where we got lucky."
In 2009, the researchers and a few angel investors formed a company, Credibility Assessment Technologies, which signed a licensing agreement with the University of Utah to commercialize the system. (The University owns the technology, since it was developed by faculty members.) A second round of experiments was conducted in 2011, this time in Colombia. The researchers were confounded by the results, however, which were very different from the first study. "It failed miserably," says Kircher. "It was like flipping a coin."
They speculated that the failure stemmed from the test subjects' lack of education. They read so poorly that the extra cognitive effort needed to just make sense of the statements overwhelmed any effects of lying. This flaw might yet turn out to be a crucial handicap.
Kircher and co. continued to labor. Eye-tracking cameras improved. The software used to plot and process the data was refined. A VC firm from Mexico invested in the project, and in 2013 the company changed its name to Converus and appointed Mickelsen as president and CEO. Soon it had a commercial product to sell.
In its current form, the EyeDetect works like this: A person sits at a computer with their face secured in front of an infrared camera, which plots the movement of their eyes as they read and respond to a series of simple true or false questions—e.g., "I am innocent of engaging in terrorist activity." The cameras measure pupil dilations up to a 10th of a millimeter, where exactly they fixate when reading the questions, and how long they take to answer. The whole thing takes 30 minutes, and is completely automated. The answers are loaded into a secure server, and uploaded to a cloud, which generates a "credibility score" in five minutes.
"It's non-intrusive, so you're not getting hooked up to a bunch of sensors," Mickelsen told Motherboard. "It's faster, cheaper, and doesn't care about your race, color, or language."
A mock crime study published this year—this time done on a literate group of 147 students at the Monterrey Institute of Technology, in Mexico—yielded results similar to the US experiments. "We had two questions," says Mickelsen. "Did the software work? We found it did. And were there cultural differences? No, there weren't." They discovered that the device could indeed work in other languages, on people of different cultures.
"We believe there is a huge market."
The EyeDetect—which has six peer-reviewed papers published—is now being used to screen employees by 135 clients across Latin America, including agencies within the governments of Mexico, Colombia, Ecuador and Peru, according to Mickelsen. An unnamed "large Middle Eastern country" has driven the development of an Arabic version. There are now over 240 different tests, covering drug use, money-laundering, ties to criminal or terrorist organizations, or any number of custom topics or combinations.
Late last year, the EyeDetect was introduced in the US. According to Mickelsen, it's now being used to vet law enforcement and correctional officers at a state level in Utah, Texas, Ohio, Louisiana, and Florida, among others. Tests are being planned to measure its ability to detect lying parolees and probationers, according to Converus. There seems to be genuine optimism that this technology might be the one to finally challenge the maligned polygraph's long-held primacy in lie detection. We've heard this before.
Scores of gadgets and techniques have appeared over the years promising to detect a lie. But at the state and local level, there exists no uniform standard to distinguish the legit from the fraudulent. Individual police departments or government agencies decide for themselves which technologies to use.
"Given what I've seen propagated over the years, it seems pretty much anything is allowed," says Maria Hartwig, a psychology professor and lie detection expert at John Jay College of Criminal Justice. "I've seen all kinds of bullshit being sold for quite a lot of money; not just gadgets, but training programs, where the claims are just ludicrous and completely out of step with the scientific literature."
Perhaps the most infamous is Voice Stress Analysis: a program that measures changes in a voice's pitch, timbre and cadence to detect deception. Police departments throughout the country bought in, desperate for a cheaper, more efficient alternative to the polygraph. It's now widely believed to be bogus; when tested in a mock crime experiment, it was actually less accurate than flipping a coin, remarkably. Sometimes cops dust if off for the express purpose of scaring suspects into a confession during an investigation. But other than that, it doesn't really do anything.
Even more galling was the Screening Passengers by Observation Techniques (SPOT) program: a six-year-long TSA pilot project that placed officers trained to detect deception using "micro-expressions" in 161 airports across the country. It was based on the work of Paul Ekman, a divisive figure among psychologists. As Hartwig, who testified during a Congressional investigation into the now-discontinued program, said to Motherboard, "It was rolled out without any kind of vetting, based on bad science. Just a huge waste of tax dollars that does nothing but inflate false confidence in security screening."
According to Mickelsen, in the coming months a "major federal agency " will be conducting its own tests measuring its EyeDetect's efficacy in screening agents in the Secret Service, Customs and Border Protection and the like. The experiments will be monitored by the National Center for Credibility Assessment (NCCA), the body responsible for vetting lie detection tech for the federal government. If the test results are consistent with previous studies, there is a very real possibility the EyeDetect could join the polygraph as one of two federally-recognized ways to detect deception.
The NCCA imprimatur, though helpful, is not necessary to use EyeDetect on Syrian and Iraqi refugees, who are neither citizens nor job applicants. Given the polygraph's unsuitability for screening large numbers for general "credibility," there exists a potentially lucrative opportunity for Converus.
EyeDetect could screen all 10,000 refugees for links to terrorist groups or organized crime in 30 days with 10 proctors, Mickelsen said. He envisions a future where the program is screening the hundreds of thousands of parolees and probationers who must regularly submit to a lie detector and government employees throughout all agencies, here and abroad. (Since this is all just talk so far, it's unclear how exactly the government will act on the results.)
"We believe there is a huge market," says Mickelsen. "We could even have an EyeDetect screening visa applicants in every embassy in the world. There is great applicability for national security across the board. Years down the road we could be a billion-dollar business."
But the literacy handicap remains. As Kircher found in Colombia, the system only works if the person can read well. We know that the pupils dilate under cognitive stress, even under oral questioning. Simply thinking about a difficult math problem makes your pupils dilate. Currently, Kircher's lab is trying to find out whether an audio examination might produce physiological changes indicative not only of cognitive strain, but of deception. This would make the system truly universal.
The impulse behind H.R. 4291 is understandable: How can we catch evildoers without inculpating innocent people? If the checkered history of lie detection technology reveals anything, it's our desperation for a shortcut to truth. We're optimists, believers that surely something will come along that will, at last, demystify deceit. Governments, private companies, and academic researchers have invested a fortune in treasure and effort to that end. Yet truth remains elusive. It probably always will.
Clarification: An earlier version of this story said Converus claims it could screen all 10,000 refugees for links to terrorist groups or organized crime in 90 days with a handful of proctors. The company now says it could do it in 30 days with 10 proctors.