We Need to Know How Often Facial Recognition Fails
Lawmakers are in a race against facial recognition companies
Welcome to General Intelligence, OneZero’s weekly dive into the A.I. news and research that matters.
One of the most serious facial recognition technology failures may never have come to light if it wasn’t for a single tweet. “I’d love to talk to you about Detroit’s facial recognition & my family,” Melissa Williams, a food blogger whose husband had been wrongfully arrested due to faulty use of facial recognition, tweeted to an outreach coordinator for the ACLU of Michigan.
Months later, the story of Robert Williams’ wrongful arrest was in the New York Times, Washington Post, and NPR. Police had run a blurry image of a Black man through their facial recognition engine, and unknown algorithmic arithmetic brought them to Robert Williams.
The New York Times indicated that Williams’ arrest was the first time a poor facial recognition match led to an arrest like this in the United States. But it may be just the tip of the iceberg.
How many people have been caught up in similar mistakes but don’t know that facial recognition led to their arrest? How many don’t know to tweet at an ACLU outreach coordinator? It’s impossible to know.
Facial recognition is pervasive across the United States and used in law enforcement agencies that range from small-town police departments to cities with billions in annual funding. It’s becoming a cornerstone of policing: Dataworks Plus, the company that sells facial recognition database software and access to algorithms in Detroit, has contracts in dozens of cities and states. Biometrics giant NEC has more than 1,000 contracts around the world.
Last week, Amazon, Microsoft, and IBM made news for pausing or stopping their facial recognition programs with police departments. (Apple, on the other hand, announced this week it was adding facial recognition to HomeKit, its software to connect your iPhone or iPad to devices like video doorbells and smart lights.) But these aren’t the real players that have their technology deployed in communities, like NEC, RankOne, and Cognitec.