Amazon's facial recognition AI confuses US politicians with convicted felons

Amazon's facial identification artificial intelligence named Rekognition, matched the pictures of 28 US Congress members, out of the total 435, with mugshots belonging to arrested criminals.

Making the matter worse, 11 of the 28 mismatches were people of colour (39 percent), including civil rights leader John Lewis. This highlighted the long-standing racial bias present among various facial recognition systems.

According to a report in The Verge, a database of 25,000 arrested people was compared to all members of the Senate and Congress and as a result, the technology failed to work efficiently on people with darker skin tones.

The American Civil Liberties Union (ACLU), which conducted this test said that the members of Congress who were falsely matched include Republicans and Democrats, men and women, of all ages, and from all across the country. The Union added that this is why Congress should join the ACLU in calling for a ban on law enforcement using facial surveillance.

Amazon recommends setting the error-tolerance for Rekognition to 95 percent for law enforcement, while the ACLU set it down to 80 percent. Furthermore, Amazon wants everyone to confirm the AI’s results with human eyes; after all, it is up to the end user to deploy any software safely and responsibly.

On Thursday when the test results were released, it provided further ammunition to critics of the nascent technology. All this has raised concerns in the society for various civil rights and immigration groups.

A spokesperson for Amazon web services questioned the ACLU’s methodology in an emailed statement. The ACLU’s test registered a match if its system had a confidence level of 80 percent, which the organisation said was the default setting for the facial recognition system Amazon offers to the public. But Amazon says it guides law enforcement officers to register matches only at confidence levels of 95 percent or above.