Police in the US used facial recognition technology to arrest a man. The tech was wrong


The rise of interest in facial recognition comes as thinning police agencies lean on a proliferation of linked public cameras to fight crime, and as the algorithms improve. — Image by macrovector on Freepik

Technology has given police vast reach to compare the faces of criminal suspects against a trove of mug shots, driver's licenses, and even selfies plucked from social media.

But a recent attempt by the Jefferson Parish Sheriff's Office to nab a high-end purse thief via facial recognition ended badly for a Georgia man who was jailed for almost a week over a false match, his lawyer says.

A detective took the algorithm at face value to secure a warrant to arrest Randal Reid, 28, in the June theft of luxury purses from a Metairie consignment shop, attorney Tommy Calogero said.

A Baton Rouge Police Department detective then adopted JPSO's identification of Reid to secure an arrest warrant alleging he was among three men involved in another luxury purse theft the same week at a shop on Jefferson Highway, court records show.

The thieves allegedly stole more than US$10,000 (RM44,050) in Chanel and Louis Vuitton purses over three days.

Local police pulled over Reid on Nov 25 as he drove on Interstate 20 in Dekalb County, Georgia, headed to a late Thanksgiving celebration with his mother, he said.

"They told me I had a warrant out of Jefferson Parish. I said, 'What is Jefferson Parish?,'" Reid said. "I have never been to Louisiana a day in my life. Then they told me it was for theft. So not only have I not been to Louisiana, I also don't steal."

Reid was booked into the DeKalb County jail as a fugitive but was let go on Dec 1, a jail official said. Calogero said JPSO detectives "tacitly" admitted the error and rescinded the July warrant.

"I think they realised they went out on a limb making an arrest based on a face," he said.

Sheriff Joe Lopinto's office did not respond to several requests for information on Reid's arrest and release, the agency's use of facial recognition or any safeguards around it. That office also denied a formal request for the July 18 arrest warrant for Reid and copies of policies or purchases related to facial recognition, citing an ongoing investigation.

Baton Rouge police also did not respond to questions about its warrant for Reid's arrest. The warrant, signed by 19th Judicial District Judge Eboni Rose, does not say how Lopinto's office identified Reid.

The case highlights the pitfalls of a technology that more law enforcement agencies are adopting across the country, even as critics point to research showing bad matches at higher rates for some populations, including Black people and women.

Some cities and states have put clamps in recent years on police use of facial recognition. But several, including New Orleans, have since backed off amid a rise in crime.

In Louisiana, little is known of the use of facial recognition outside of New Orleans, where the City Council this year rolled back a two-year ban and set some rules.

Police in New Orleans say it can only be used to generate leads and that officers must get approval high up before lodging a request through the Louisiana State Analytic and Fusion Exchange in Baton Rouge. All possible matches also must undergo a peer review by other facial recognition investigators under the new city rules.

Police tool or privacy scourge?

Elsewhere in Louisiana, there is no regulation. A state bill to restrict use of facial recognition died in 2021 in committee.

The libertarian Pelican Institute for Public Policy argued in favour of legal guardrails. The Louisiana Sheriffs' Association was against the bill. The association's executive director, Michael Ranatza, argued that more study was needed.

Loren Lampert, executive director of the Louisiana District Attorneys' Association, also spoke against the bill.

Lampert said last week that the technology is used by police only to develop suspects in cases where there are none, and that any match must be corroborated, like with a fingerprint. "I think all agree that there are concerns that must be addressed as to accuracy and the potential for misuse," he said.

Ranatza said he didn't know how many Louisiana sheriffs now use the technology.

"It's a tool. It helps us to identify a suspect. I've not been made aware of where it's used exclusively" to secure someone's arrest, he said last month. "It was always an assistance to law enforcement to establish probable cause."

That portrayal minimises the technology's expansive reach and potential for abuse in government hands, argued Chris Kaiser, advocacy director for the ACLU of Louisiana.

"We're not just looking at a few photos. You're searching a library of thousands and potentially millions of photographs, potentially nationwide and even outside of the country," he said. "It's apples and oranges."

Kaiser said no police agencies will admit to booking a suspect based on an algorithm alone.

"They will always say this is for an investigative lead," he said. "But there really isn't any protection behind that."

Public records obtained by the Southern Poverty Law Center for 2021 show that Lopinto's office was among several local, state and federal agencies to request facial recognition analyses through the state intelligence hub.

The records show three requests from JPSO in 2021. State Police did not immediately respond to a request for similar records for 2022.

Whether Lopinto's detectives also run facial recognition searches outside of the state clearinghouse is uncertain.

Widening tentacles

Reports issued by the Fusion Center in 2021 indicate that it employs two facial recognition providers: Clearview AI and Morphotrak.

The web site for Clearview AI boasts "the world's largest facial network," with tens of billions of images "sourced from public-only web sources, including news media, mugshot websites, public social media, and many other open sources."

Morphotrak is now Idemia, a French company that touts hundreds of government clients for its facial recognition products.

The rise of interest in facial recognition comes as thinning police agencies lean on a proliferation of linked public cameras to fight crime, and as the algorithms improve.

A 2021 study by the National Institute of Standards and Technology found all the top algorithms can identify airport passengers with photos in the system more than 99% of the time on the first appearance before a camera. But that study and others also found continuing bias in differing rates of false positives and negatives by race and gender.

Some critics have argued that the technology is so perilous in government hands that it should be banned. The Electronic Privacy Information Center argued this year that facial recognition is "inherently dangerous," enabling "comprehensive public surveillance."

"It's a powerful surveillance tool that can easily be expanded without people's knowledge," said Jeramie Scott, senior counsel with EPIC.

Doppelganger blues

If Reid was a false match for an algorithm, he was a pretty good guess, Calogero said.

He described Reid, who is Black, as a spitting facial image of a man caught on camera in June entering numbers from a stolen credit card at the register of Second Act on Metairie Road. The thieves left with US$7,500 (RM33,037) in purses, according to a sheriff's incident report.

Baton Rouge police claim the same three men stole a Chanel bag priced at US$2,800 (RM12,334) through a bogus credit card purchase at Swap Boutique. Baton Rouge detective Samuel Stafford wrote that the men in JPSO's case "appear to be the same subjects" involved in the Swap heist.

Stafford also secured an arrest warrant in the case for a 21-year-old New Orleans man. It's unclear whether that man was arrested in the case, however.

Reid said he sat in jail fearing he could lose his job as a transportation analyst and land a pair of felony convictions for crimes he didn't commit.

"Not eating, not sleeping. I'm thinking about these charges. Not doing anything because I don't know what's really going on the whole time," he said. "They didn't even try to make the right ID."

Differences, such as a mole on Reid's face prompted JPSO to rescind the warrant, said Calogero. He estimated a 40-pound difference between Reid and the purse thief he saw in surveillance footage. The culprit's "flabby arms" were a clear tell, he said.

"Police could have checked his height and weight or made an effort to speak to him or asked to walk through his house to look for evidence. He would have complied," Calogero said.

"There are 300 million people in this country. All of us have someone who appears identical to us." – The Times-Picayune | The New Orleans Advocate/Tribune News Service

Follow us on our official WhatsApp channel for breaking news alerts and key updates!
   

Next In Tech News

Reducing your Netflix footprint: A guide to sustainable streaming
Opinion: I hope you don’t see your passwords on this list
Telegram, long a lawless chat app, is now aiding a content crackdown
Being online brings mental health benefits for over-50s, study shows
Advanced AI chips cleared for export to UAE under Microsoft deal, Axios reports
US TikTok content creators warn followers to find them on Instagram, YouTube
Drones offer an unparalleled�view of the biggest wildfire risks
Teaching a robot surgery through videos
'Sexting' scam: Criminals impersonate cops in US to steal money from dating app users
FBI releases tips for customers to avoid scams when holiday shopping

Others Also Read