Detroit police challenged facial recognition flaws, bias

Detroit police challenged facial recognition flaws, bias

A black man who says he was wrongly arrested for facial recognition technology mistakenly identifying him as a suspected shoplifter calls for a public apology from Detroit police. And for the department to refrain from using the controversial technology.

Robert Williams’s complaint is a rare challenge from someone who has not only experienced an erroneous facial recognition hit, but has also discovered that he was responsible for his subsequent legal troubles.

The Wednesday complaint filed on behalf of Williams alleges that his photo of a Michigan driver’s license – kept in a state library of images – was incorrectly marked as a likely match with a suspected shoplifting. Investigators had scanned grainy surveillance camera footage of an alleged 2018 robbery at a Shinola watch store in downtown Detroit, police records show.

That led to what Williams describes as a humiliating January arrest for his wife and young daughters on their front yard in the Detroit suburb of Farmington Hills.

“I can’t really put it into words,” Williams said in a video message about the daytime arrest that left his daughters crying. “It was one of the most shocking things that ever happened to me.”

The 42-year-old automotive worker, backed by the American Civil Liberties Union, demands a public apology, final dismissal of his case, and Detroit police to scrap the use of facial recognition technology. Several studies have shown that current facial recognition systems are more likely to make mistakes in identifying people with dark skin.

The ACLU complaint said the Detroit police “thoughtlessly relied on flawed and racist facial recognition technology without taking reasonable steps to verify the information provided.” Calling the resulting investigation “sloppy and incomplete,” the agents involved were “rude and threatening,” the department joined forces to respond to public inquiries for relevant documents.

Detroit police and Wayne County prosecutors did not immediately reply to a request for comment by email on Wednesday.

DataWorks Plus, a South Carolina company that supplies facial recognition technology to Detroit and Michigan State Police, was also not immediately available for comment.

Police records indicate that the case started in October 2018, when five expensive watches went missing in the flagship store of Detroit-based luxury watchmaker Shinola. A loss prevention officer later reviewed the suspect’s video footage of a black man wearing a St. Louis Cardinals baseball cap.

“Video and photos were sent to Crime Intel for facial recognition,” said a brief police report. “Facial Recognition came back with a hit” – for Williams.

At the top of the facial recognition report, prepared by Michigan State Police, was a bold, capitalized warning that the computer’s finding should be treated as an investigator, not as a likely reason for arrest.

But Detroit detectives then showed a six-photo lineup, including Williams to the loss prevention officer, who positively identified Williams, according to the report. It took months for the police to issue an arrest warrant, and several more before calling Williams at work and asking him to come to the police. It is not clear why.

Williams said he thought it was a joke. But they came to his house shortly after, took him away in handcuffs and held him at night. It was during his interrogation the following day that it became clear to him that he was not properly identified by facial recognition software.

The investigative officer looked confused, told Mr. Williams that the computer said it was, but then acknowledged that “the computer must have misunderstood,” says the ACLU complaint.

Prosecutors later dismissed the case, but without prejudice – meaning they could pursue the case again.

The case is likely to spark a move in Detroit and around the US to protest police brutality, racial injustice and the death of George Floyd at the hands of Minneapolis police. Detroit activists have brought reforms to the city’s mayor and chief of police, including turning down the police and ending the use of facial recognition.

Providers of police facial recognition systems often point to research showing that they can be accurate when used correctly under ideal conditions. A review of the industry’s leading facial recognition algorithms by the National Institute of Standards and Technology found that they were more than 99% accurate when matching high-quality headshots with a database of other frontal poses.

But if you try to identify a face using a video feed, especially using ceiling-mounted cameras commonly found in stores, accuracy can drop sharply. Studies have also shown that facial recognition systems don’t perform equally well in race, gender and age – work best for white men and have potentially harmful effects on others.

Concerns about bias and increasing surveillance of police practices after Floyd’s death led tech giants IBM, Amazon and Microsoft to announce earlier this month that they would stop selling facial recognition software to the police, at least until Congress can issue guidelines for its use. Several cities, led by San Francisco last year, have banned the use of facial recognition by municipal authorities.

Copyright 2020 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or distributed without permission.

.