The tech is freeing up Customs and Border Protection employees—once they learned to trust it.
It’s been difficult for Customs and Border Protection insiders to accept that facial recognition technology is now better at identifying humans than their human eyes are, but that’s also a primary reason why the agency is increasingly deploying it, Director of Policy Michael Hardin said at an event in Washington Wednesday.
“This was really hard for us to sort of admit to ourselves,” Hardin said on a panel hosted by the Information Technology and Innovation Foundation. “But the reality of that is there are some things that machines are better at and there are some things people are better at. Right now, it’s clear that machines are better at matching a photo to a person—and that’s shown up in all of our research.”
Hardin explained that, like many things across CBP and the Homeland Security Department, their use of biometrics and facial recognition largely stems from the aftermath of 9/11. Once follow-up reports revealed that many of the hijackers employed document forging methods to enter and exit the United States using various identities, the technology was introduced into the realm of immigration and border-crossing.
Today, Hardin said the agency’s main use for facial recognition is to confirm that people are who they say they are as they move in, out and around the nation.
“It’s a huge advantage for us now, not just because the machine can perform better than the human in the actual matching, but also because it frees up the person to do other law enforcement activities in a small amount of time, which is really all they have,” he said.
Benji Hutchinson, vice president of federal operations at the information technology company NEC Corporation, also differentiated between how facial recognition has come to be used across law enforcement versus how Hardin and others are implementing it across immigration services. Hutchinson said it’s used in family reunification to identify matches of children that have been kidnapped, it helps dispel people who are wrongfully convicted and it supports officials in developing investigatory leads.
He also noted that Homeland Security and other components have recently launched a great deal of research around how to use the technology to review images of sexually exploited children on the dark web to try to identify the victims and their assailants.
“That’s another area that’s very cutting-edge and very new, but it’s a promising use case,” he said.
Hutchinson said he’s seen “dramatic increases in the accuracy of the technology” over the last five years and noted that tests from the National Institute Standards and Technology revealed error rates as low as 1%. He said tests with high-performing algorithms and massive datasets of more than 5 million faces have error rates that are even lower.
“I like to remind people that NIST tested four algorithms in the late ‘90s, and last year they did over 150—we are talking a dramatic increase in competition,” he said. “We do understand that it’s never 100 percent and that’s why we spend millions of dollars a year to keep fixing the algorithms and making them better.”
Hardin also noted that CBP funded research at NIST to help them better understand bias and ensure their algorithms are working effectively.
“We wrote NIST a big check,” he said.
The policy director added that although the agency “does not collect data on race,” they’ve conducted some of their own internal studies including evaluating flights that are boarded biometrically instead of with boarding passes and flights that go to other sides of the world, in an effort to assess differentiation in the technology’s performance.
“We haven’t found anything consistent in any particular direction yet, but that doesn’t mean that we don’t take it seriously and we are not looking at it really seriously,” Hardin said.