MIT researcher warns Amazon about bias in facial recognition software



[ad_1]

An MIT researcher warned Amazon CEO Jeff Bezos last month that the company's controversial facial recognition software is the most inaccurate about women and individuals of color.

In a letter of June 25, Joy Buolamwini, who also founded the Algorithmic Justice League to fight coded biases in algorithms, stated that skewed technology could exacerbate existing racial prejudices in the police.

Amazon sells its facial recognition software, known as "Rekognition", to law enforcement agencies in the country.

While the software displayed extremely low error rates for men with paler skin, Buolamwini than women of all skin colors and individuals of both sexes with darker skin face much higher rates of inaccuracy.

Of all the male faces tested, only 0.14 percent were wrongly identified. Lighter skin faces of both sexes were incorrectly identified in approximately 2% of cases. Women, meanwhile, were misidentified more than 16% of the time and darker faces of both sexes were poorly identified in 13% of cases.

She also warned that the direct nature of her tests, performed on still images, far more likely to produce accurate results than on potentially fuzzy images in a range of different types of lighting in the real world .

"Given what we know, it is irresponsible to use these systems, and I support a moratorium on the use of facial recognition by the police," Buolamwini told The Hill. "I also support the call for federal regulations on facial recognition technology because, unlike Canada, the US has no federal law on biometric data."

Buolamwini used the same methodology in his facial recognition test of Amazon. 19659009] previous analysis of facial recognition software from IBM and Microsoft, who also both showed higher rates of misidentification of dark-skinned women.

His study on Rekognition of Amazon is a more complete version of the one published by the ACLU on Thursday, which also found models of prejudice against people of color. The ACLU study was conducted on 535 members of Congress, while the Buolwimini study focused on a dataset of more than 1,200 faces.

The ACLU, along with Amazon employees, Amazon shareholders and other groups, call on the company to sign facial recognition contracts with the forces of order and to establish a moratorium on facial recognition

[ad_2]
Source link