Police Facial Recognition: System Named a Living Man Dead

Full Episode Transcript

A computer told police that an eighteen-year-old man was dead. It named him as someone else entirely, a criminal from a different city. The only reason anyone caught the mistake? The man it named was alive, out on bail, and very much walking around.

Trusted by Investigators Worldwide

Run Forensic-Grade Comparisons in Seconds

Detailed facial comparison reports. Results in seconds.

Get Started

7-day refund guarantee**

This happened in Ludhiana, India

This happened in Ludhiana, India. Police say two alleged shooters died in an encounter with officers. They uploaded photos of the dead into a system called the Punjab Artificial Intelligence System. The software got one man right. The second? It flagged a young man named Gurpreet Singh as someone called Sandeep, because their faces looked alike. If you've ever thought a face-scan just confirms who someone is, this story is about you. Because here's the question that hangs over the whole thing, what happens when police treat a computer's guess as a fact?

Let's start with what actually went wrong in Ludhiana. The system didn't fail completely. It correctly matched one suspect. That's what makes this so slippery. One right, one wrong, same system, same moment, same incident. The error only surfaced when outside information reached police. Someone learned the man the software named was alive. Human detective work caught what the machine missed. For anyone who's ever trusted a confident answer on a screen, that should stick with you.

Now, why do these systems get faces wrong in the first place? A study out of M.I.T. found the error rate for dark-skinned women was nearly thirty-five percent. For lighter-skinned men? Under one percent. And researchers at N.I.S.T., the U.S. government's testing lab, found false matches up to a hundred times more likely for Black and Asian faces than for white men. A hundred times. So the tool doesn't just make mistakes. It makes them unevenly, depending on who you are.

And this isn't only happening overseas. A grandmother in Tennessee spent nearly six months in jail after software misidentified her. Six months. For a match that was wrong. Reporters say she's at least the ninth American this has happened to. Back in twenty-twenty, police in Detroit arrested a man based only on a facial recognition result. He was innocent. The city later paid him three hundred thousand dollars. The next time you unlock your phone with your face, remember, the same basic idea can put the wrong person in a cell.

The Bottom Line

There's a technical reason these errors pile up. Facial matching works best with a clear, straight-on photo in good light. Feed it a side angle, a blurry frame, or a grainy A.T.M. still, and accuracy falls apart. Error rates for those poor images can climb above twenty percent. That's the gap between a lab demo and a real crime scene.

But here's what most people get backwards. A facial match was never supposed to be proof. It's a lead, a starting point, not an ending point. The technology can't legally be used as probable cause. The failure isn't the algorithm. It's the humans who stop investigating the moment the screen says "match."

So let's bring it home. A computer named a living man as a dead criminal, and only luck and legwork caught it. These systems miss more often on women and people of color. And a match is just a guess that needs checking, not a verdict. Whether you carry a badge or just carry a phone, the lesson is the same. A face on a screen is a question, not an answer. The full story's in the description if you want the deep dive.

Police Facial Recognition: System Named a Living Man Dead