Amazon created its “Rekognition” face-recognition software to be used by law enforcement agencies to flawlessly identify people. So the American Civil Liberties Union (ACLU) decided to test it.
The ACLU assembled a database of 25,000 publicly available photos of people. Then it showed Rekognition random photos of people and asked the software to find photos of those same people in the database. Rekognition mistakenly matched the faces of 28 members of Congress with police mug shots of known criminals.
In a more serious vein, the same software more often mismatched the faces of people of color. In one case, it matched OJ Simpson’s photo with that of a white male with a mustache. These mistaken tendencies led the ACLU to point out that police who trust the software could focus suspicion on a totally innocent person.
In another pilot test, the Orlando, Florida, police department has been using Rekognition to match “face grabs” from city surveillance cameras with the department’s own mug shot database. No word on accuracy yet.
Facial recognition won’t sweep the world in a wave. It will begin in limited settings, where the software has limited choices, such as in businesses needing high security but with a relatively small number of employees.
Because of the dangers of misidentification, growth should proceed cautiously over the long term.