Since September, Lincoln police have used new state Department of Motor Vehicles facial recognition software to help them identify 23 possible suspects in crimes ranging from shoplifting to burglary to illegal gun purchases.
The use of that tool — one that civil liberty advocates still have concerns about despite evolving technology — was formalized last month when the Lincoln City Council approved a memorandum of understanding with the Nebraska Department of Motor Vehicles.
“This is an important tool for us,†Police Chief Teresa Ewins told the council. “And to make it perfectly clear to everyone, this is not the way in which we go and arrest anyone. You need a lot more than a hit on facial recognition. It is a tool.â€
The use of facial recognition software to identify potential suspects isn’t new to LPD.
In 2013, the City Council approved an agreement with the DMV to allow police to run searches as part of criminal investigations — but didn’t allow officers to use findings as the sole basis for an arrest.
In 2019, police considered buying video processing software that also included facial recognition capabilities, but the DMV purchased upgraded software and officials decided entering into the agreement with the state was a better option, said Erin Sims, a former police sergeant who now supervises the department’s forensic lab.
Jared Minary, an LPD video technician, said in reality, police rarely used the old DMV software because it was really only effective when the image police had was of a person looking directly at a camera — which rarely happens with surveillance videos.
But the new software also used by other law enforcement agencies is more adept at analyzing surveillance photos. For instance, Sims said, it can take an image at an angle and build the second side of the face based on algorithms. Still, LPD will only run searches if the image it has is good enough quality to successfully search. LPD doesn't run searches based on police sketches or "look-alike" photos of celebrities — cases noted in some national studies of use of the technology.
Chad Marlow, senior policy counsel with the ACLU in New York, said the software produces more misidentifications with brown and black faces — a group already over-targeted by law enforcement.
Also, he said, using DMV photos as a database means all residents who have a driver’s license are potential suspects.
“You’re either allowed to have your privacy or drive. Pick one,†Marlow said. “I don’t think that most Nebraskans would appreciate having their feet held to the fire like that simply because they want to get to work at the grocery store.â€
Listen now and subscribe: | | | |
Another problem, Marlow said, is “automation bias,†which is the tendency for people to trust computers. That bias means an officer is likely to be hesitant to reject all the possibilities suggested by a search.
One of the recommendations from a Georgetown University report is to require “double-blind confirmation†by two analysts independently to conclude the same photo is a possible match.
And local law enforcement officials stress that a possible match doesn't mean a person is a suspect.Â
The technology is better described as “facial similarity,†said Minary and Sims, because it uses facial measurements from an image to match with photos that have similar measurements. Police typically get multiple possible matches to examine.
“Once the computer looks at algorithms and gets a top list of candidates, it’s no longer the computer's job to figure out who the suspect is, it’s the officer’s,†Minary said. “It’s just like a Crime Stopper’s tip — you can’t take it at face value. You have to validate.â€
Once officers decide on a possible match, they must build a case on other evidence.
Case in point: Surveillance from a Walmart store recently captured images of two people stealing groceries and plants as they scanned items in the self-checkout. Those images were run through the DMV software, which came up with a number of potential matches, Minary said.
Investigators found two images that could be matches. The man identified as a possible suspect had a tattoo on his arm that matched the one captured in the surveillance video. He was contacted, admitted to stealing the items, was arrested and convicted, Minary said. The officer had enough information about the woman in the Walmart video, who is wanted for five similar crimes, to issue an arrest warrant.
That’s one of the 23 cases in Lincoln the new DMV’s facial recognition software helped solve since September. Of those, 12 resulted in arrests, Minary said, though not all those arrests occurred because of the matches. Eleven other cases that came up with potential matches are still under investigation.
About half the cases were shoplifting, but also included burglaries, larcenies from vehicles where people stole credit cards then tried to use them, someone giving false information to try to buy guns, “porch pirate†cases where delivered packages were stolen from porches and the theft of an ATM machine, Minary said.
The software the DMV uses is made by the same company that makes the fingerprinting software police use, he said. Unlike fingerprints, facial recognition images aren’t used as evidence in court.
Minary and a few other LPD employees will be trained to use the new software as part of the memorandum of understanding approved by the council, which will be good for four years with the option of renewing it for two more four-year terms.
Spike Eickholt, the government liaison with ACLU Nebraska, said while the technology is a convenience for police, the concerns about inaccuracy and privacy invasion means his office is committed to making sure LPD keeps its word that the technology won’t be used to “surveil, target or harass innocent people.â€
Ewins, who is from San Francisco — the first major police department to ban the software’s use — said she is aware of the concerns and wants to make sure LPD is protecting people’s Fourth Amendment rights.
“It’s all about checks and balances,†she said. “We are and will always be aware the software is not a panacea of identifying someone.â€