The Detroit PD — or at least a few of its investigators — managed to be the absolute worst at using facial recognition tech. Despite being told by their tech provider that a “match” was never to be considered probable cause for an arrest on its own, multiple PD investigators decided a match — no matter how iffy — was all they needed to start locking people up.
One of those people was Robert Williams, who was arrested in the driveway of his home because officers claimed grainy, dark footage from a jewelry store robbery showed he was the most likely suspect. The final part of this thoroughly incomplete “investigation” was performed by the PD’s facial recognition tech, which used a non-ideal image to serve up a match one detective (who has been sued twice so far) treated as actual probable cause.
Here’s the image they were working with:
Not great! Not even really acceptable! The face — the thing needed most for facial recognition — was in shadows, not clearly defined, and of low enough quality you would think the AI would have served up nothing more than a shrug.
Nonetheless, detectives Daniel Bussa and Steve Posey decided this was enough to go after Williams, who was arrested and, following his exoneration, sued the detectives for violating his rights.
That lawsuit has now netted Williams a $300,000 payout from Detroit residents who had nothing to do with this bogus arrest. But that’s how city funding works and that’s why when government agencies get things wrong, they rarely feel compelled to prevent the same thing from happening again. Here are a few more details on this part of the settlement, courtesy of Emma Camp at Reason:
Williams filed a lawsuit against Detroit police in 2020. Last month, the case was settled, with Williams receiving a $300,000 payout, according to the Associated Press. As part of the settlement, the Detroit Police Department also agreed to conduct an audit of all cases “in which facial recognition technology was utilized to generate an investigative lead that was followed by an arrest or the issuance of an arrest warrant.”
“We are extremely excited that going forward there will be more safeguards on the use of this technology with our hope being to live in a better world because of it,” Williams told journalists last month. “Even though what we would like for them to do is not use it at all.”
While I am pleased to see someone secure compensation for being wronged by the government (even if it will be paid by citizens, rather than the government employees involved), the bigger win is the other part of the settlement, which provides far more of a deterrent to future abuse of this tech than spending other people’s money to escape a lawsuit.
The other part of the settlement alters how the Detroit PD handles facial recognition tech matches, as well as how it conducts its virtual “lineups” when talking to eyewitnesses. That part of the settlement will force the PD to play by the rules set by the tech provider, which means investigators won’t be able to consider search matches probable cause on their own. In addition, any matches served up by the AI will no longer take prominence in photo arrays presented to eyewitnesses. Instead of a “six pack” of photos, investigators will have to present photos one at a time, including a separate photo of the suspect that hasn’t been derived from the facial recognition tech. Investigators will also have to disclose to eyewitnesses (and courts) that the match was derived from an automated search, rather than pretend these photos were just part of a random selection.
Finally, let’s not forget that a system capable of serving up false positives is just as capable of pitching false negatives to investigators. Relying solely on AI search results — as Detroit PD detectives did in multiple cases — compounds the problem. Not only are innocent people being arrested but guilty parties are going undetected. Facial recognition can work if it’s a small part of much larger and better vetted investigation process. What it should never be expected to do is act as a stand-alone probable cause for arrests.
