Facial Recognition Was Never Designed to See All Faces

Our Take

Prateek Raj's video does a solid job walking through the receipts on facial recognition bias — Joy Buolamwini's Gender Shades audit, the federal government's own NIST numbers, and the Robert Williams wrongful arrest that turned a statistical error rate into 30 hours in a jail cell. None of this is new to anyone who's followed the fight against face surveillance, but it's worth restating plainly: these systems were trained, tested, and sold by companies that didn't prioritize accuracy for everyone equally, and the people who paid for that gap were disproportionately Black and Asian faces.

We cover ALPRs, not facial recognition, but the pattern is identical. Flock Safety and its peers built a nationwide camera network first and worried about accountability, error rates, and misuse later — if at all. Just like commercial face-matching algorithms, license plate readers generate false positives, feed data into opaque databases, and get used by police and, increasingly, private actors, with almost no public oversight of who's watching, how accurate the system actually is, or what happens when it gets it wrong. The Robert Williams case isn't a facial-recognition problem, it's a surveillance-infrastructure problem, and ALPR networks are quietly building the same kind of unaccountable machine.

Credit to Prateek Raj for laying out the data clearly. If you want to see how far this infrastructure has already spread into your own neighborhood, check our camera map, and if you're ready to push back on it locally, our take-action page has concrete next steps.