Amazon's Facial Recognition System Identifies 28 US Congressional Members as Criminals

The American Civil Liberties Union (ACLU) announced Thursday that it had tested Amazon's facial recognition system, Amazon Rekognition, and that the results were shocking because the system was identifying incorrectly 28 US Congressmen Arrested for a crime, raising serious concerns among lawmakers about the technique while Amazon actively buys this system to law enforcement agencies across the United States, which is simply not good enough.

According to Jack Snow, a lawyer for the Civil Liberties Union: "The organization has loaded the Amazon system with a database containing 1945,000 [25] criminal records [1945900] for criminals and perpetrators Legal offenses downloaded from a public source. By pbading the official photos of all members of Congress 535 members at the Rekognition asking them to match one of these images can match the database and ended up matching images of 28 members, 28 failed 6 members because they had dark skin including MP John Lo A Democrat from Georgia, an African-American activist in the field of civil rights

In its response, Amazon said: "It's about defining faces during the test." However, these results highlight the risks to which individuals Can not cope if the police use this technology to catch criminals.

The American Civil Liberties Union and other civil rights groups urge Amazon since May 2018 to stop selling Rekognition to law enforcement, where civil rights groups have reported the use of the Rekognition system by the forces of order. Florida and warned that this tool can be used to unfairly target migrants and people by their color.

A study by the American Civil Liberties Union showed that the Amazonian system exhibited a racial bias now that the Amazonian system did not allow members to identify 39% because they had the dark skin, whereas they represented only 20% of Congress. Dark-skinned people suffer disproportionately. "

Researchers have shown that racial prejudice is a well-established problem with facial recognition systems. In February, researchers at the Mbadachusetts Institute of Technology discovered that Facial recognition of Microsoft, IBM and FaceScript took longer to identify In people with darker skin tones than whites, and in the June evaluation of the Amazon Rekognition system, the researchers found that the system's algorithms were biased in the same way that the system could correctly identify Oprah Winfrey

Three of the Democrats identified in the American Civil Liberties Union – Senator Edward Markey, Attorney Luis Gutierrez and Representative Marc de Saulniere – have sent a letter to the CEO of Amazon Jeff Bezos to express his concern and interest to question the accuracy of the tool and its use. Before the law enforcement authorities.

The American Civil Liberties Union stated that it had paid only $ 12.33 to compare the Amazon system between the official photographs of each member of the House of Representatives and the Senate in exchange of a database of 25 000 photographs of the authors.

Facial recognition technology is already widely used in China by the police, and there are a number of startups actively seeking to apply this technology. "We are always excited about the opportunity to use image and video badytics to be a good driving force in the world," said a spokeswoman for Amazon's cloud services division, citing technical badistance to find missing children and prevent crime. Human and do not make any final decisions. She commented on the problem of badigning indicators to her system during the test, noting that the ACLU was using the 80% confidence limit for the system – that is, the likelihood that the Rekognition system find a match and you can adjust it to the required level of accuracy.

"The 80% confidence rate was appropriate for identifying things, animals or other situations of use. social media, not individuals.If the system is used by law enforcement agencies, we order clients to set a minimum of 95% or higher. "The idea that human support brings a better performance to the system is thus confirmed:" People often rely too much on technological systems as they were infallible, "said Jeremy Scott, director the local monitoring project of the Electronic Privacy Information Center.In 2009, for example, the San Francisco police arrested a woman and held her under the threat of a gun after the reader registration plate has incorrectly identified his car Corresponding but

Even if facial recognition works well, putting it in the hands of law enforcement agencies still raises concerns. that we try to avoid here, it's collective supervision, "said Scott.

While the ACLU study shows another aspect of FAS errors, it may have a better chance of To have a real impact. "The asp The most important aspect of this test is that it has been made personal to members of Congress, "said Alvaro Bedoya, executive director of the Center for Privacy and Technology at Georgetown University. The American Civil Liberties Union has attracted the attention of many lawmakers about these mistakes. "

Privacy advocates claim that at least the use of facial recognition by law enforcement agencies must be restricted until racial prejudice is over." be corrected and guaranteed, and even in this case, they must limit their use.

"The technology that has proven very imprecise between people on the basis of their skin color is unacceptable in the work of the police in the 21st century, "said Bidoya.