Amazon’s facial recognition technology mistook 28 members of Congress for arrested individuals when the American Civil Liberties Union ran lawmaker faces against a database of 25,000 mug shots, the group announced Thursday.
In its test of Amazon’s Rekognition technology, the software also disproportionately misidentified people of color, the ACLU said. Forty percent of the members wrongly identified were people of color, while that same demographic makes up just 20 percent of Congress.
Rekognition identified women much more accurately, the test found, only missing on Rep. Norma Torres.
The ACLU is calling for a moratorium on the federal use of Rekognition until government officials have considered the implications, particularly on minority communities.
“People of color are already disproportionately harmed by police practices, and it’s easy to see how Rekognition could exacerbate that,” the ACLU said in a statement.
For example, the use of the software could wrongly identify a suspect as having a previous arrest, biasing an officer before an encounter even begins, the group said.
The Congressional Black Caucus also wrote a letter to Jeff Bezos about the unintended consequences the software could have on African Americans, immigrants, and protesters.
However, the test used the default setting of an 80 percent similarity score between two photos to produce a match, an Amazon Web Services spokesperson said. The online retail giant and web services company said it would recommend law enforcement a 95% threshold.
Rekognition’s findings were not meant to be determining but rather an aid in making informed decisions, the spokesperson said.