Facial Recognition: India's Top Court Weighs Wrong Matches

Seven people have been confirmed wrongly accused because a facial recognition system pointed at the wrong face. The Innocence Project says six of those seven were Black. Now India's Supreme Court is asking how a wrong match gets caught before it lands on a real person.

If you've ever been photographed in a crowd, this

If you've ever been photographed in a crowd, this story is about you. Your face can be captured at a protest, a train station, or a street festival without you ever knowing. And if a computer says you look like someone else, you may be the one answering for it.

Here's what happened. A member of India's upper house of Parliament asked the Supreme Court to examine how Delhi Police used facial recognition and biometric surveillance at a recent student protest at Jantar Mantar. The Court agreed to take it up and issued notice. Police say any match gets physically verified before they act. So the question running through this whole story is simple. When a machine flags you, what does checking it actually look like?

Start with a question from the bench. Justice Joymalya Bagchi asked whether an ordinary security camera is the same thing as an A.I. system that can identify you out of a giant database. It sounds like a small point. It isn't. A regular camera records what happened. A facial recognition system tries to tell police who you are.

That difference matters because of how often the second kind gets it wrong. The article says these systems make more mistakes on certain groups. Those include Black people, women, East Asians, and older people. In one set of research, a false match for darker-skinned people happened about a hundred times more often than for others in the comparison. The article describes that as two orders of magnitude.

For an investigator, that means a match can't be

For an investigator, that means a match can't be treated as an answer. For everyone else, it means the person most likely to be mistaken for a stranger may depend on who they are.

Now take the police defense. Police say a human checks every match before anyone is stopped. That sounds reassuring. But the article points out that a check is only as good as the person doing it.

Picture an officer who's just been told the computer found a suspect. That officer now expects to see a match. That's confirmation bias, the habit of seeing what you've already been told to see. A second look can end up as a rubber stamp.

That's why the article draws a hard line. Accuracy isn't confidence. A system can sound sure and still be wrong. So a match should only ever be a starting point, never the conclusion.

That's true whether you're a detective or a person

And that's true whether you're a detective or a person standing in a crowd who never asked to be scanned.

There's one more thread. Reports on the case say the Court is looking at two issues. One is facial recognition of protesters. The other is the data held by private companies. Lawyers in the case lean on India's landmark twenty-seventeen privacy ruling, known as Puttaswamy. The question underneath is what privacy you can still expect in a public place.

The Court isn't asking whether the technology exists. It's asking who catches the mistake, and how.

A tool that compares faces in one case, using photos an investigator chose, is different from a system that scans everyone and sorts them afterward.