Facial Recognition and Wrongful Arrest: When a Match Becomes Proof

It was a working day like any other when the police arrived at the door. Alvi Choudhury, twenty-six years old, a software engineer of Bangladeshi origin, was logged in from the family home in Southampton, the unremarkable rhythm of a remote developer's morning unfolding around him: meetings, messages, lines of code. Then officers were on the step. They handcuffed him. They told him he was suspected of a burglary, a theft of roughly three thousand pounds' worth of property, that had taken place in Milton Keynes. Milton Keynes is a hundred miles away. Choudhury had never been there.

What had placed him at the centre of an investigation in a town he could not have found on a map was not a witness, not a fingerprint, not a stray possession left at the scene. It was a piece of software. Thames Valley Police had run footage of the genuine suspect through a facial recognition system and the algorithm had returned a name: his. From that single computational suggestion flowed everything that followed, the cuffs, the cell, the nearly ten hours in custody before he was released at two o'clock in the morning.

Choudhury looked nothing like the man in the footage. By his own account the suspect appeared roughly a decade younger, with lighter skin, no facial hair, a larger nose, different eyes, different lips. He said as much. He asked the officers, in a question that ought to chill anyone who believes a custody suite is a place of careful judgement, whether the image on the screen looked anything like him. According to his account, they laughed. Thames Valley officers, he later said, conceded that they had known he was not the suspect after comparing the footage with his photograph. They proceeded anyway.

This is the part that should not be skated over. The failure here was not, in the first instance, the machine's. Machines err. The deeper failure was that a chain of human beings, each with the authority to stop, looked at a face that did not match and kept going regardless,…