He said supermarket staff were unable to explain why he was being told to leave, and would only direct him to a QR code leading to the website of the firm Facewatch, which the retailer has hired to run facial recognition in some of its stores. He said when he contacted Facewatch, he was told to send in a picture of himself and a photograph of his passport before the firm confirmed it had no record of him on its database.
“One of the reasons I was angry was because I shouldn’t have to prove I am innocent,” Rajah said. “I shouldn’t have to prove I’m wrongly identified as a criminal.” He described the incident as feeling “quite like Minority Report, Orwellian”.
He said while doing his normal shop, he was approached by three members of the store’s staff, one of whom appeared to affirm that he was the person pictured on a device they had. It is understood the Facewatch system flagged someone else who had entered the store, and staff mistook Rajah for him.
Rajah was concerned some form of permanent record implying he had been involved in criminality might have been created on Facewatch’s system. Eventually, the firm told him he was not on its database and referred him back to Sainsbury’s.
I remember reading something last year about how some not most facial recognition software has racist biases because the models are primarily trained on Caucasian people. Based on just the name, I’m going to assume Rajeh is not white, and may be a victim of that bias. Granted, it shouldn’t be used at all, so it’s splitting hairs more than anything.
IIRC they found that even with balanced training data facial recognition models just do worse on darker-skinned people. Something about cameras picking up less contrast on the skin, meaning there are fewer easily-identifiable facial features it can pick up from an image.
I remember reading something last year about how some not most facial recognition software has racist biases because the models are primarily trained on Caucasian people. Based on just the name, I’m going to assume Rajeh is not white, and may be a victim of that bias. Granted, it shouldn’t be used at all, so it’s splitting hairs more than anything.
IIRC they found that even with balanced training data facial recognition models just do worse on darker-skinned people. Something about cameras picking up less contrast on the skin, meaning there are fewer easily-identifiable facial features it can pick up from an image.