Facial recognition led to wrongful arrests. That’s why Detroit is making changes.

In January 2020, Robert Williams spent 30 hours in a Detroit jail because facial recognition technology identified him as a criminal. The match was false and Mr. Williams filed a lawsuit.

On Friday, as part of a court settlement over his wrongful arrest, Mr. Williams received a pledge from Detroit police to do better. The city adopted new rules governing police use of facial recognition technology that the American Civil Liberties Union, which represented Mr. Williams, said should be the new national standard.

“We hope it moves things in the right direction,” Mr Williams said.

Mr. Williams was the first person proven to have been wrongfully arrested based on faulty facial recognition. But he was not the last. Detroit police arrested at least two other people based on failed facial recognition, including a woman who was charged with auto theft while eight months pregnant.

Law enforcement agencies across the country are using facial recognition technology to identify criminals whose offenses have been caught on video. In Michigan, the software compares an unknown face to faces in a database of mugshots or driver’s license photos. In other jurisdictions, police use tools like Clearview AI, which scours photos from social media sites and the public internet.

One of the most important new rules passed in Detroit is that images of people identified through facial recognition technology can no longer be shown to an eyewitness in a photo lineup unless there is other evidence linking them to the crime.

“The ‘take a photo and line it up’ approach will be over,” said Phil Mayor, attorney for the ACLU of Michigan. “This settlement takes the Detroit Police Department from the best-documented abuser of facial recognition technology to a national leader in adhering to strict rules in its use.”

Police say facial recognition technology is an effective tool for solving crimes, but some cities and states, including San Francisco, Austin, Texas, and Portland, Oregon, have temporarily banned its use over privacy concerns and racial bias. Stephen Lamoreaux, chief of information technology for the Detroit Criminal Investigation Division, said the police department is “very interested in using technology in a meaningful way for public safety.” Detroit, he claimed, has “the most stringent policies in the country right now.”

How it goes wrong

Mr. Williams was arrested following a crime in 2018. A man stole five watches from a downtown Detroit boutique while being filmed by a security camera. A theft prevention company provided the footage to Detroit police.

A comparison of the man’s face with driver’s license photos and mug shots produced 243 photos, sorted by system confidence, that were of the same person in the surveillance video, according to documents disclosed as part of Williams’ lawsuit. An old driver’s license photo of Mr. Williams was ninth on the list. The person conducting the search deemed him the best match and sent a report to a Detroit police detective.

The investigator inserted Mr. Williams’s picture into a “six-pack photo lineup” — photos of six people in a grid — which he showed to the security officer who had provided the store’s surveillance video. She agreed that Mr. Williams was the closest match to the man in the boutique, and this led to the warrant for his arrest. Mr. Williams, who had been sitting at his desk at an auto parts company when the watches were stolen, spent the night in jail and his fingerprints and DNA were taken. He was charged with retail fraud and had to hire a lawyer to defend himself. Prosecutors eventually dropped the case.

He sued Detroit in 2021, hoping to force a ban on the technology so others don’t suffer his fate. He said he was angered last year when he learned that Detroit police had charged Porcha Woodruff with car theft and robbery after a bad facial recognition match. Police arrested Ms. Woodruff as she was getting her children ready for school. She has also sued the city; the case is ongoing.

“It’s so dangerous,” Williams said, referring to facial recognition technology. “I don’t see any positive use in it.”

The new rules

Detroit police are responsible for three of the seven known cases in which facial recognition has led to a wrongful arrest. (The other cases occurred in Louisiana, New Jersey, Maryland and Texas.) But Detroit officials said the new controls will prevent further abuse. And they remain optimistic about the technology’s potential to solve crimes, which they currently use only for serious crimes such as assault, murder and home invasions.

James White, Detroit’s police chief, blames “human error” for the wrongful arrests. His officers, he said, relied too heavily on the clues provided by technology. Their judgment was flawed, not that of the machine.

The new directive, which came into force this month, is intended to help with this. Under the new rules, the police can no longer show a person’s face to an eyewitness based on facial recognition alone.

“There has to be some kind of secondary, corroborating evidence that is unrelated before there is sufficient justification for the confrontation,” said Mr. Lamoreaux of the Detroit Criminal Investigation Department. For example, the police would need location information from a person’s phone or DNA evidence – so more than just a physical resemblance.

The department is also changing the way it conducts photo lineups. It is implementing what is known as a double-blind sequencing process, which is considered a fairer way to identify a person. Instead of presenting a witness with a “six-pack,” an officer – who does not know the main suspect – presents the photos one at a time. And the lineup includes a different photo of the person than the one the facial recognition system found.

Police must also disclose that a face search took place, as well as the quality of the image of the face being searched for – how grainy was the surveillance camera? How well can the suspect’s face be recognized? – as a poor quality image will produce less reliable results. They must also disclose the age of the photo displayed by the automated system and whether there were other photos of the person in the database that did not show up as a match.

Franklin Hayes, Detroit’s deputy police chief, expressed confidence that the new measures would prevent future misidentifications.

“There are still a few things that could go wrong, for example with identical twins,” Hayes said. “We can never say never, but we believe this is our best course of action yet.”

Arun Ross, a computer science professor at Michigan State University and an expert on facial recognition technology, said Detroit’s policy is a great starting point and other agencies should adopt it.

“We don’t want to trample on the rights and privacy of individuals, but we also don’t want crime to get out of hand,” Ross said.

How much does it help?

Identifying eyewitnesses is a delicate undertaking and police have concluded that cameras and facial recognition are more reliable than imperfect human memory.

Police Chief White told local deputies last year that facial recognition technology helped “take 16 murderers off the streets.” When asked for more information, police department officials did not provide details about those cases.

To demonstrate the department’s success with the technology, police officials instead played surveillance video showing a man spraying gasoline and setting it on fire at a gas station. They said he was identified using facial recognition technology and arrested that same evening. He later pleaded guilty.

The Detroit Police Department is one of the few agencies that monitors its facial recognition searches and submits weekly reports on their use to an oversight board. In recent years, it has averaged more than 100 searches per year, with about half of those searches producing potential hits.

The department only tracks how often it receives a tip, not whether that tip is confirmed. However, as part of the settlement with Mr. Williams – who also received $300,000, according to a police spokesman – it must conduct a review of its facial recognition searches dating back to when the technology was first used in 2017. If it identifies additional cases in which people were arrested with little or no supporting evidence other than a facial match, the department must notify the appropriate prosecutor.

Molly Kleinman, director of a technology research center at the University of Michigan, said the new protections sound promising, but she remains skeptical.

“Detroit is an extraordinarily monitored city. There are cameras everywhere,” she said. “If all this surveillance technology really did what it says it would, Detroit would be one of the safest cities in the country.”

Willie Burton, a member of the Board of Police Commissioners, an oversight body that approved the new guidelines, called them “a step in the right direction,” but he continued to oppose police use of facial recognition technology.

“The technology just isn’t there yet,” Burton said. “One false arrest is one too many, and three of them in Detroit should be a red flag to stop.”