Amazon ’s controversial face recognition package , holler Rekognition , misidentified more thantwo dozen members of Congress as people arrested for crime . The delusive identifications were made when the ACLU of Northern California tasked Rekognition with match photos of all 535 member of Congress against 25,000 publicly available mugshot photos . The trial run be the ACLU just $ 12.33 to do .
In total , Rekognition misidentified 28 member of Congress , number them as a “ match ” for someone in the mugshot picture . 11 of the misidentified member of Congress were people of color , a extremely alarming disparity . Tests have shown that face recognition isroutinely less accurateon darker skinned people and cleaning lady . For Congress as a whole , the error pace was only five pct , but for non - white members of Congress , the computer error rate was 39 percent .
Among those misidentified were six members of the Congressional Black Caucus , whowrote an open letterto Amazon CEO Jeff Bezos in June after the ACLUreleased inner documentsconcerning the purpose of Rekognition by police .
Amazon has encourage practice of law enforcement agencies to adopt Rekognition , with police in Orlando latterly opting tocontinue their pilotof the technical school . In a blog post , the ACLU of Northern California explained how misidentification in the hands of constabulary enforcementcould be deadly , peculiarly for people of color :
If law enforcement is using Amazon Rekognition , it ’s not hard to think a police police officer come a ‘ match ’ indicate that a person has a premature hold back - weapon arrest , biasing the officer before an showdown even begins .
In a financial statement , Amazon suggested ACLU ’s results with the software “ could belike be meliorate by following good practice , ” saying the company recommend law enforcement agencies set a confidence verge of at least 95 pct when deciding if there ’s a match . Of course , there ’s no effectual requirement for police force to do so . In the UK , law in South Walesrecently defended using boldness acknowledgement software with a staggering 90 percent false positive charge per unit .
Amazon ’s instruction is included in full below :
We have see customers employ the image and telecasting depth psychology capability of Amazon Rekognition in ways that materially profit both company ( for instance prevent human trafficking , inhibiting child exploitation , reunite miss children with their family , and building educational apps for children ) , and organizations ( enhancing security through multi - factor certification , obtain icon more well , or prevent package larceny ) . We remain delirious about how picture and picture analysis can be a driver for near in the domain , include in the public sector and law enforcement .
With regard to this recent trial of Amazon Rekognition by the ACLU , we imagine that the termination could plausibly be improved by follow good practices around setting the confidence threshold ( this is the portion likeliness that Rekognition found a match ) used in the trial run . While 80 % sureness is an satisfactory brink for photograph of hot bounder , electric chair , animals , or other societal media use cases , it would n’t be appropriate for identifying individuals with a reasonable grade of certainty . When using facial recognition for law enforcement bodily function , we guide client to position a eminent threshold of at least 95 % or higher .
Finally , it is deserving remark that in real earth scenario , Amazon Rekognition is almost exclusively used to help constrict the field and grant human to efficiently reexamine and take options using their mind ( and not to make fully self-directed decision ) , where it can avail find lost baby , curb human trafficking , or forestall crimes .
Face Recognition
Daily Newsletter
Get the best tech , scientific discipline , and culture news in your inbox day by day .
News from the future , fork up to your nowadays .
You May Also Like
