How Reckless Cops Relying On Questionable Facial Recognition Tech Can Destroy Lives

The use of facial recognition far outpaces its proven track record. Prone to false positives and negatives, especially when it comes to anyone else but white males, the tech continues to make inroads with the law enforcement community which has never seen a black man it can’t prosecute for crimes he didn’t commit.

That’s the case here. Randal Reid, a Georgia resident, was picked up by Georgia law enforcement relying on a tip passed on to them by Louisiana law enforcement. The origin of the so-called “tip” was the Jefferson Parish Sheriff’s Office, which used its facial recognition tech to turn Randal Reid into the prime suspect in a string of luxury purse thefts.

There were several problems with this assumption, starting with the tech and ending with the Baton Rouge PD, who decided to “adopt” this so-called tip and pass it on to neighboring agencies. The tech employed by the Sheriff’s office said search results should not be considered probable cause. The law enforcement agencies involved in this wrongful arrest either ignored that warning or were never apprised of this fact.

Either way, it ended in the arrest of Reid — someone who didn’t actually match the description and had never traveled to the Louisiana towns where the alleged fraud had taken place. Reid spent nearly a week in jail before the agency that started this whole debacle stepped in and “rescinded” its obviously bogus warrant.

The New York Times has done an in-depth investigation of AI’s failure to properly identify criminal suspects, focusing on Reid’s life-destroying encounter with facial recognition tech. This is how it started. And how it started is enough to end life as they know it for anyone falsely accused of a crime.

[Reid’s] parents made phone calls, hired lawyers and spent thousands of dollars to figure out why the police thought he was responsible for the crime, eventually discovering it was because Mr. Reid bore a resemblance to a suspect who had been recorded by a surveillance camera. The case eventually fell apart and the warrants were recalled, but only after Mr. Reid spent six days in jail and missed a week of work

The arrest warrant makes no reference to the tech used to misidentify Reid. All it says is that a detective watched the surveillance video and the suspect caught on camera “appeared to match the description the suspect” when paired with info from Louisiana’s Department of Motor Vehicles database. But that non-admission of facial recognition tech is likely just another layer of law enforcement deception.

Unfortunately for the law enforcement agencies, the information withheld from the judge approving the arrest warrant has leaked out as Reid continues to seek justice for his false arrest. And it began with a Jefferson County Sheriff’s officer making the mistake of referring to this as a “positive match,” suggesting tech capable of cross-referencing CCTV footage with law enforcement databases had been used.

The other law enforcement entities that participated in this false arrest made similarly revealing statements, albeit not to a judge who might have had questions about the tech used and/or the quality of the video image being used to run searches. But the Jefferson County Sheriff’s Office did most of the lying.

Andrew Bartholomew, the Jefferson Parish financial crimes detective who sought the warrant to arrest Mr. Reid, wrote in an affidavit only that he had been “advised by a credible source” that the “heavyset black male” was Mr. Reid. 

Oh, really? Would Mr. Reid be able to cross-examine this “credible source?” Would he be able to ask a judge to determine whether this “source” was indeed “credible?” Of course not. The “credible source” was an algorithm — one that remains unproven for several reasons. But the most important reason is that the Sheriff’s Office is unwilling to inform the courts that it’s using this tech to seek warrants.

And it gets worse. Rather than utilizing a more respectable facial recognition tech purveyor, the Sheriff’s Office decided to go with the cheapest, easiest, shadiest option available.

The Sheriff’s Office has a contract with one facial recognition vendor: Clearview AI, which it pays $25,000 a year. According to documents obtained by The Times in a public records request, the department first signed a contract with Clearview in 2019.

Even Clearview is smart enough to tell law enforcement customers that matches alone shouldn’t be considered probable cause for searches or arrests.

The company’s chief executive, Hoan Ton-That, said an arrest should not be based on a facial recognition search alone.

But clicking “ok” on a dialog box before seeking a warrant isn’t a deterrent. (Yanking contracts when cops abuse Clearview’s search results would be, but Clearview definitely isn’t going to do that.) Cops aren’t going to police themselves, so outside parties working with cops need to do this job for them. Unfortunately, the desire for market share often overwhelms the desire to protect people from government abuse.

So even if Clearview says (as it does in this article) that it has “tremendous empathy” for those falsely accused as the result of its tech, it’s not going to stop selling access to its AI or its billions of facial images scraped from the internet.

The end result is the abomination seen here: cops arresting the wrong man for the crime of using a fake credit card to “purchase” a designer purse. But it’s even worse than that. The warrant affidavit suggests an innocuous search of the state DMV database to identify the suspect. That may have happened but it apparently wasn’t until after the Sheriff’s Office ran screenshots pulled from CCTV footage against Clearview’s multi-billion image database that it was able to finger the wrong person for the job.

There’s even more tech involved here, each bit of it designed to reduce the friction between the real world and things cops want to do. Clearview’s facial recognition appears to be implicated here — a mass surveillance tool that turns the open web into a law enforcement playground.

On top of that, there’s the e-warrant service the sheriff’s office uses, known as CloudGavel. For around $40,000 a year, the Sheriff’s Office can send warrant requests 24/7 to judges who only need to click a couple of buttons to set everything in motion. While it’s probably preferable to getting judges out of bed or interrupting them during numerous government holidays, it also makes it that much easier to apply a rubber stamp to get back to Inbox Zero.

This particular warrant was e-signed at 4:28 pm on July 18, 2022 — well within normal government business hours. But it’s a lot easier to get something signed when you don’t have to be confronted in person about the deficiencies of your warrant affidavit. And it’s a lot easier to rubber stamp something when it’s almost quittin’ time. I would normally say I’m not suggesting either of these things happened here, but in this case I’m going with the other: this looks like government interests aligning to give cops a questionable warrant and a judge a quicker start to their evening.

Law enforcement is still in the clear at the moment. But Reid’s life is a mess. A week in jail awaiting the supposed thief with Louisiana law enforcement officers so desperate to catch this purse fraudster that they omitted information from their warrant request. A week with no job, no income, and the added insult/injury of having his car impounded, just because. No matter the outcome, the officers involved will likely keep their jobs. Even if their cars had been impounded, there are always plenty of transportation options at work. No one gets hurt but the little people. And all because a computer said “Maybe?” and investigators chose to read it as “Definitely.”