Concerns over how the government deploys facial recognition tools were on full display during a U.S. House of Representatives hearing examining the Department of Homeland Security's use of biometric technologies held Wednesday morning.

The Committee on Homeland Security hearing lasted a little over two hours and broached topics related to privacy, data security, and even discrimination against women and people of color that could result from limitations inherent in facial recognition technology. By the time the proceedings drew to a close, it seemed unlikely that many of those concerns were inching any closer to a resolution.

“The American people deserve answers to those questions before the federal government rushes to deploy facial recognition technology further,” said committee chairman and Rep. Bennie Thompson, D-Mississippi.

The hearing came less than a week after the Washington Post revealed that state driver's license databases across the country were being used by the Federal Bureau of Investigation (FBI) and Immigration and Customs Enforcement (ICE) for facial recognition efforts. Specifically, photos from those databases were being accessed to help not only to identify suspects in a crime, but also potential witnesses and victims as well.

Thompson referenced the Washington Post story directly in his opening remarks, stating that ICE was scanning through the drivers licenses of millions of Americans without their consent. Congressional representatives at the hearing also frequently referenced last month's hack of a U.S. Customs and Border Protection (CBP) subcontractor, which compromised images that were being used as part of an ongoing facial recognition pilot that tracks people going in or out of the U.S.

John Wagner, CBP's deputy executive assistant commissioner, who served as a witness for the hearing, was unable to specify how much time had passed before they were notified by the subcontractor about the breach, only venturing that it was a “significant amount of time.”

Wagner also indicated that CBP was in the process of adding audit controls to all of its systems to ensure that a portable media drive cannot be connected to external networks and used to extract information. He said the subcontractor involved in last month's breach had taken images off of CBP cameras and put them into their own network.

Throughout the hearing, those kinds of security risks were consistently balanced against the need for some of kind of technological assist in optimizing the security and travel process.

Both Wagner and fellow witness Austin Gould, assistant administrator for requirements and capabilities analysis for the Transportation Security Administration (TSA), cited that the use of facial recognition can be used to improve boarding rates at airports. According to Gould, passenger volumes grow at approximately four percent annually. He pointed to 4.8 million passengers and crew that were processed over the 4th of July weekend alone.

Gould also addressed questions about a pilot program at Hartsfield Jackson International Airport in Atlanta that uses facial recognition to verify a passenger's identity prior to boarding a flight. The potential for discrimination was a recurring concern raised throughout the hearing.

“In a review of our data, we are not seeing any significant error rates that are attributable to a single demographic,” Gould said. 

Still, the problem of discrimination has not yet been eliminated from facial recognition technology. Charles Romine, director of the Information Technology Laboratory at the National Institute of Standards and Technology, said that while the best facial recognition algorithms can approach a range of 99.7 percent accuracy, it's unlikely that there will ever come a point where that performance is identical for every single demographic across the board.

According to Romine, facial recognition performance also depends greatly on the quality of the image being used. For example, images captured through a windshield or where people are walking and not facing the camera directly can produce poor results.

“It's still true: garbage in, garbage out,” he said.