Grandmother Jailed, Detroit's 96% False Positive Rate, and the Facial Recognition Reckoning
How many more grandmothers need to sit in jail cells before we admit facial recognition policing is fundamentally broken?
The latest victim: an innocent woman in North Dakota, misidentified by AI and jailed for months in a fraud case she had nothing to do with. It's becoming a depressingly familiar pattern.
<> Former Detroit Police Chief James Craig acknowledged that using facial recognition alone would yield misidentifications 96% of the time./>
Let that sink in. Ninety-six percent. If you deployed any other technology with a 96% failure rate in law enforcement, heads would roll. But somehow facial recognition keeps getting a pass, despite mounting evidence of its catastrophic unreliability.
Detroit: The Facial Recognition Hall of Shame
While details of the North Dakota case remain sparse, Detroit has become ground zero for facial recognition disasters. The city has managed to falsely arrest at least four Black individuals using this flawed technology:
- Robert Williams was arrested in January 2020 for allegedly stealing a watch. Wrong person, wrong crime. His case resulted in the first settlement requiring actual policy changes—imagine that, making cops do additional investigative work before ruining someone's life.
- Porcha Woodruff, a pregnant nursing student, spent 11 hours detained for a carjacking she didn't commit. Prosecutors eventually dismissed the case for "insufficient evidence"—legal speak for "we screwed up royally."
Between Williams's arrest and settlement, Detroit managed to rack up two more false arrests using the same broken system. It's almost impressive in its consistency.
The Algorithm Bias Nobody Wants to Fix
Here's the dirty secret everyone knows but pretends doesn't matter: facial recognition algorithms are primarily trained on white faces. This isn't some accidental oversight—it's a fundamental flaw baked into the technology from day one.
The bias is so pronounced that people of color face dramatically higher misidentification rates. Yet police departments keep deploying these systems, apparently content to let marginalized communities bear the cost of their technological incompetence.
Hot Take: Ban It All
Facial recognition in law enforcement needs to be banned. Not regulated. Not "improved with better training data." Banned.
The technology's defenders love to argue about accuracy improvements and bias mitigation. They're missing the point entirely. Even if facial recognition achieved 99% accuracy tomorrow, that still means 1 in 100 identifications would be wrong. In a country that processes millions of criminal cases annually, that's thousands of innocent people swept up in the system.
And let's be honest about the "improvements." We've been hearing about bias fixes for years, yet innocent people keep getting arrested. The North Dakota grandmother is just the latest casualty in an ongoing disaster.
<> The technology's bias is particularly pronounced for people of color because algorithms are primarily trained on datasets of white faces./>
The Real Cost
Behind every misidentification statistic is a human being. A grandmother who spent months in jail. A pregnant woman interrogated for 11 hours. A man arrested in front of his family for a crime he didn't commit.
These aren't "growing pains" or "edge cases." They're predictable outcomes of deploying fundamentally flawed technology in high-stakes situations.
Police departments have access to decades of investigative techniques that don't involve algorithmic guesswork. Maybe it's time they remembered how to use them.
Until then, expect more grandmothers behind bars and more settlements quietly sweeping the mess under the rug. The facial recognition reckoning is here—the only question is how many more innocent people will pay the price.
