Amazon facial recognition falsely matched 28 in Congress with mugshots in ACLU experiment
U.S. Congress includes 28 individuals matched to arrest photos by Amazon’s facial recognition technology, according to a blog post by the American Civil Liberties Union (ACLU).
In a demonstration of the system’s accuracy and limitations, the ACLU ran images of the 535 members of Congress against a database of 25,000 publicly available mugshots with default match settings, for the cost of $12.33. With a false match rate for this experiment of just over 5 percent, Rekognition found incorrectly identified members of Congress ranging in skin color, gender, age, and party affiliation, including civil rights legend Rep. John Lewis (D-Ga.).
The ACLU notes that the Congressional Black Caucus (CBC) sent Amazon CEO Jeff Bezos a letter (PDF) expressing concerns with the use of the technology by law enforcement shortly after the ACLU and other privacy groups touched off a controversy about the use of Rekognition by U.S. police with a letter to the company.
“It is quite clear that communities of colour are more heavily and aggressively policed than white communities,” the CBC writes. “This status quo results in an oversampling of data which, once used as inputs to an analytical framework leveraging artificial intelligence, could negatively impact outcomes in those oversampled communities.”
Almost 40 percent of the false matches in the ACLU test were for people of color, nearly double the 20 percent congressional share.
Ever AI CEO Doug Aley, a veteran of the field of artificial intelligence and machine learning, believes that with this level of accuracy, human oversight should be required.
“Any technology used by the police can be abused,” Aley told Biometric Update in an email. “The most important thing is for law enforcement to understand that no system is foolproof. After making an identification, there still needs to be human intervention and the presumption of innocence until you have manually verified a suspect. Regulation and training is needed for sure, but we also need to understand that having these tools will indeed save lives in the long run.”
“If law enforcement is using Amazon Rekognition, it’s not hard to imagine a police officer getting a ‘match’ indicating that a person has a previous concealed-weapon arrest, biasing the officer before an encounter even begins. Or an individual getting a knock on the door from law enforcement, and being questioned or having their home searched, based on a false identification,” the ACLU says in the blog post.
“An identification — whether accurate or not — could cost people their freedom or even their lives.“
The post goes on to describe an incident in which an incorrect match by an automatic license plate reader led to an unnecessarily aggressive confrontation between a police officer and an elderly black woman. It says that Oregon has begun using Rekognition to build a mugshot database without publc debate, and warns that surveillance with facial recognition could harm First Amendment rights.
The ACLU concludes by calling on Congress to enact a moratorium on the use of facial recognition by U.S. law enforcement.
Amazon Web Services President of Worldwide Public Sector Teresa Carlson responded to a wave of criticism this week saying that the company will make no changes to the way it markets and licenses facial recognition technology to the U.S. government and law enforcement.