Facial Recognition: WA Police Scan 1.4M Faces, Arrest 122

Western Australian police pointed cameras at crowds and scanned almost one point four million faces. That trial ended with one hundred twenty-two arrests. Police haven't said how many people the system missed.

If you've ever walked through a shopping area or a

If you've ever walked through a shopping area or a train station, your face may have been part of a count like this one. You didn't sign anything. You probably didn't see a sign. That can feel unsettling, and it's reasonable to feel that way. This was Australia's first police trial of live facial recognition. Cameras scanned faces in real time and compared them against a list of wanted people. Police say faces that didn't match were deleted right away. So the question is simple. What does one hundred twenty-two arrests actually tell us?

Start with the math. Out of nearly one point four million faces, one hundred twenty-two led to arrests. That's a tiny sliver of the crowd, less than one hundredth of one percent. It shows the tool can pick a few people out of a huge flow of strangers. It doesn't show the tool was right every time. An arrest count measures scale. It doesn't measure accuracy.

Facial recognition can fail in two ways. It can miss a wanted person. Experts call that a false negative. Or it can flag the wrong person. That's a false positive. Researchers at the University of Reading report that leading algorithms miss fewer than one percent of wanted people. The wrong-person rate is lower than half a percent. Those sound small. But apply a small rate to one point four million faces, and small stops being small. Police in Western Australia haven't published how many people the system missed. They also haven't broken down how it performed across different groups of people.

That second gap matters for real people

That second gap matters for real people. The Innocence Project reports at least seven confirmed cases in the U.S. where Deepfake 1 In 15 Australians Hit And Familiar Voices Lie led to a misidentification. Six of those seven people were Black. Seven people. Wrongly named by a machine. If a trial doesn't report results by group, nobody outside the program can tell whether it treats everyone the same.

Courts have noticed a related problem. According to legal scholars, no court has examined facial recognition under rigorous scientific review and then accepted it as proof of identity. The method hasn't earned general acceptance in the scientific community. That sounds like a bombshell. But it mixes up two different things. Using a match as a starting point for an investigation happens today. Putting the raw machine output in front of a jury as proof is another matter, and judges rightly scrutinize it. The Western Australia trial didn't produce convictions from matches alone. It produced leads that officers had to follow up.

Public opinion leans toward the technology. Surveys show close to ninety percent of people support police using it to investigate crimes and find missing persons. Police say it helps protect vulnerable people. Those are real benefits, and they deserve weight.