As an unsuspecting woman prepared her two daughters for school one February morning, she was shocked to find six armed police officers at her door. They requested she step outside and promptly arrested her for carjacking. The woman, compliant, was stunned.
Is This for Real?
The woman had no idea what was happening or why they thought she was connected to a carjacking. At eight months pregnant, she was in no condition to pull off a carjacking and would never be compelled to participate in that kind of activity anyway. Even so, police handcuffed her and took her to jail. She spent 11 hours detained, locked in a holding cell, and being questioned relating to a crime she knew nothing about. Finally, she was released on a $100,000 bond. Then she went to a local hospital, where she was treated for dehydration. One long month later, the charges against her were dropped.
Why Was This Woman a Suspect?
The evidence leading to the woman’s arrest began with facial recognition technology (FRT). Videotape from the crime scene was run against a criminal mugshot database, which came up with the innocent suspect. Her face was there due to a 2015 arrest for driving under an expired driver’s license. Police included the woman’s photo in a photo lineup that was shown to the victim, who identified her as his assailant. And therein lies the problem: When a person appears similar to the perpetrator, facial recognition technology and humans can both make the same identification mistakes.
Can FRT Be Trusted?
There is general agreement that facial recognition technology is extremely accurate under model conditions. However, in most cases captured in video surveillance from things like convenience store security cameras or traffic cameras, conditions are appreciably less than ideal. Indirect angles, shadows, and aging have adverse effects on the reliability of FRT. One study found that the technology’s accuracy rate at sporting venues was between 36% and 87%. That is a substantially lower rate than in, for example, an airport, where security cameras are placed at boarding gates in well-lit, eye-level conditions. It brings into question how law enforcement can be so confident in FRT results.
Race and FRT
Another serious problem with this technology is the problem related to race and FRT. Research is clear: The rate of error is over a third higher for dark-skinned people than for light-skinned people, with black women experiencing the highest error rates when attempts to identify them were tested. It makes one wonder why law enforcement would rely on FRT at all. Continue reading