The FBI is building a facial recognition database that could potentially contain the face of every American, but it barely clears the accurate return rate requirements.
The FBI has been using facial recognition technology since 2010. Its Next Generation Identification-Interstate Photo System contains millions of photos of Americans, which FBI officers use during criminal investigations. The Government Accountability Office found that the database, which cost $1.2 billion to create, correctly returns matches 86 percent of the time. The National Institute of Standards and Technology requires facial recognition returns to be correct 85 percent of the time.
Diana Maurer, director of homeland security and justice issues at the Government Accountability Office, said that the FBI also does not check for false positives in their results. Maurer testified at a House Committee on Oversight and Government Reform hearing on facial recognition technology.
“Instead of relying on books of mug shots from the Hill Street Blues era, law enforcement can use CSI-era computers to nearly instantly identify someone from a grainy photo,” Maurer said. “However, pictures of millions of Americans are being searched by the FBI, which is why attention to accuracy and privacy is so important. We found the FBI needs to do a better job on both fronts.”
Rep. Jason Chaffetz, R-Utah, chairman of the House Oversight and Government Reform Committee, expressed doubts with the FBI’s ability to protect the identities contained in its photo database. He cited the 2015 Office of Personnel Management data breach that exposed millions of peoples’ identities and fingerprints.
“What scares me is that the FBI is proactively trying to collect everybody’s face. I don’t trust the Federal government. I don’t believe they can keep all this information secure,” Chaffetz said. “The technology will show us, the statistics will show us, the bigger the database, the more difficult it is for facial recognition to get it right.”
The accuracy of NGI’s return rate was only one point of concern regarding the FBI’s database. Millions of people who have never been arrested populate the database. According to Chaffetz, 80 percent of photos in the FBI database are noncriminal entries, meaning the photos were gathered from drivers’ licenses and passports.
Alvaro Bedoya, executive director of Georgetown Law’s Center on Privacy and Technology, said that one out of two Americans are in the FBI’s criminal recognition network. He also said he knows a college student who was arrested for peaceful civil disobedience and now covers her face at protests.
“Facial recognition lets law enforcement identify someone from far away and in secret,” Bedoya said. “This technology raises some serious questions. Is it a good idea to give the government so much power?”
GAO also found that the FBI has been slow in updating privacy impact assessments for this system. The Department of Justice created a PIA for the FBI in 2008, as mandated by the E-Government Act whenever agencies develop technologies that collect personal information. However, the FBI did not update the PIA when the system underwent significant changes in 2015.
Rep. Stephen Lynch, D-Mass., agreed with Chaffetz, stating that the number of people who are registered unwittingly on the FBI’s facial recognition database was an invasion of privacy. He said the FBI should have an ombudsman sift through the database and remove people who did not have any criminal offenses on their record.
“I have zero confidence in the FBI keeping this in check. Innocent people should not be on this database,” Lynch said. “This is really Nazi Germany here. They had meticulous files on individuals. I see little difference in the way people are being tracked. This is corrosive of our very liberty.”
Another issue surrounding facial recognition technology is how different algorithms categorize different races.
Facial recognition devices do not perceive all races with the same degree of accuracy. Jennifer Lynch, senior staff attorney for the Electric Frontier Foundation, said that facial recognition technology misidentifies African-American and Hispanic people more frequently than it misidentifies white people.
Rep. Elijah Cummings, D-Md., said technology’s disproportionate effect on African Americans is a serious issue. Cummings lives in Baltimore, which has been the site of protests since Freddie Gray died in police custody in 2015. Cummings cited a 2012 report from the FBI revealing that several of the leading algorithms used in these recognition systems were 5 to 10 percent less accurate when used on African Americans.
“I have seen the impact of certain police tactics with regard to African-American males. Having been an African-American male for 66 years, I have a lot of concern about this,” Cummings said. “According to these reports, if you’re black, you’re more likely to be subjected to this technology and this technology is more likely to be wrong. That’s a hell of a combination.”