Face-Off: Amazon’s facial recognition tool really messed up this time

Amazon’s (AMZN) facial recognition tool Rekognition misidentified 28 Congress members as illegal suspects. The tool was tested by the American Civil Liberties Union (ACLU) which, along with other civil rights advocates, have been rallying against Amazon demanding that the company refrain from selling the application to law enforcement agencies.

In a test that cost just over $12, the ACLU compared the photos of all the US House and Senate members against a database of 25,000 mugshots. The results turned up matches for 28 Congress members with an accuracy level of around 80%. The matches also highlighted people of color more than others, giving strength to the arguments by civil rights groups that face-surveillance methods would unfairly target minorities and immigrants.

Civil rights activists have been arguing for long that methods of surveillance using facial recognition technology could be used unethically against minority communities, immigrants, activists or just about anyone with a viewpoint that is different from that of the government. ACLU wants Congress to prohibit law enforcers from using the technology on the grounds of it being faulty, biased and dangerous.

Amazon meanwhile stood by its product and blamed incorrect settings for the unflattering test results. Amazon said the 80% setting was more suited for the identification of objects and not for human beings and when being used in law enforcement for identifying people, a setting of 95% or higher was recommended.

Despite the criticism, Amazon believes the Rekognition tool can be used for crime prevention, tracking of missing individuals as well as several other worthy purposes. Facial recognition is popular in China for legal use with several companies investing in it significantly.