AI’s facial recognition failures: Three times crime solving intelligence got it wrong

Law enforcement’s use of artifical intelligence-driven facial recognition puts everyone into what one expert called a “perpetual police line-up,” and studies show it’s more likely the finger will be pointed at the wrong person if they’re Black or Asian. 

“Whenever they have a photo of a suspect, they will compare it to your face,” said Matthew Guariglia, from the nonprofit digital rights group Electronic Frontier Foundation, told the BBC. “It’s far too invasive.”

The technology’s use in police investigations boomed in recent years, particularly after the Jan. 6 Capitol riot. 

Twenty out of 42 federal agencies that were surveyed by the Government Accountability Office in 2021 reported they use facial recognition in criminal investigations.

AI MIGHT HAVE PREVENTED BOSTON MARATHON BOMBING, BUT WITH RISKS: FORMER POLICE COMMISSIONER

But the artificial intelligence algorithms in the technology falsely identified African American and Asian faces 10 to 100 times more than White faces, according to a 2019 study by the National Institute of Standards and Technology. 

Since then, several Black men have been arrested for crimes they didn’t commit. These are three recent cases.

Robert Williams, Detroit, 2020

Police used a blurry surveillance photo of a suspect in two jewelry store robberies, and the algorithm matched it to Williams’ driver’s license.

The case sparked widespread criticism of the technology and a lawsuit from the American Civil Liberties Union.

ACLU FILES COMPLAINT OVER FACIAL RECOGNITION ARREST

“We know that facial recognition technology threatens everyone’s privacy by turning everybody into a suspect,” Phil Mayor, senior staff attorney for the ACLU, told Fox 2 Detroit in April 2021.

“We’ve repeatedly urged the Detroit Police Department to abandon its use of this dangerous technology, but it insists on using it anyhow. Justice requires that DPD and its officers be held accountable.”

Watch Vintra’s demo:

The charges against Williams were eventually dropped, but he was held in jail for 30 hours, according to the ACLU lawsuit. 

This was the first reported wrongful arrest due to an incorrect facial recognition match, but Detroit police made the same mistake in 2019. 

Michael Oliver, Detroit, 2019

Oliver was 25 when he was arrested during a traffic stop and accused of grabbing a phone from a teacher who was recording a fight outside a school, then breaking it.

Except Oliver was nowhere near the scene; he was working at the time. The arrest cost him his job, and he said in previous interviews that he thought he was going to be wrongly convicted.

REGULATION COULD ALLOW CHINA TO DOMINATE IN THE AI RACE, EXPERTS WARN: ‘WE WILL LOSE’

“I’ve got a son, I’ve got my family, I’ve got my own little house, paying all my bills, so once I got arrested and I lost my job, it was like everything fell, like everything went down the drain,” Oliver told Wired during an interview last year.

The charges were ultimately dropped after about a year, and Oliver spent 10 days in jail. He sued Detroit police in federal court. 

The same Detroit detective was involved in Oliver’s and Williams’ arrests, according to Wired, which cited court documents. 

MEET THE 72-YEAR-OLD CONGRESSMAN GOING BACK TO COLLEGE TO LEARN ABOUT AI

Randal Quran Reid, Atlanta, 2022

Reid was arrested by police in Atlanta the Friday after Thanksgiving 2022 on theft warrants from Baton Rouge, Louisiana, a state he’s never visited. 

It took days for Reid to learn he was accused of using stolen credit cards to buy designer handbags. 

After a week of phone calls to family and lawyers, Reid was ultimately released and the charges were dropped. He said he’s considering a lawsuit.

“Thousands of dollars for something I didn’t do,” Reid told The New York Times.

 

The case was spotlighted by a recent New York Times story, which reported that the use of facial recognition was never included in official documents, a practice that is reportedly becoming more prevalent. 

Baton Rouge police haven’t officially said the misidentification was because of the use of facial recognition, or that the technology was used, but “a person with direct knowledge of the investigation” confirmed its use to The New York Times.