Microsoft Tackles Race and Gender-Based Errors in Facial Recognition

Microsoft Tackles Race and Gender-Based Errors in Facial Recognition

Facial recognition software has a come a long way in a short time. While the actual development is extremely important, testing also plays a major role in getting things off the ground. This brings us Microsoft and its Face API.

Problems With Race and Gender-based Recognition

The company announced it made some big improvements on Face API. Earlier this year, the software was the target of criticism regarding its error rate identifying the gender of people of color. The study noted a rate of 2008-percent.

The software had more success with lighter skin tone with a perfect score on lighter male faces. The trouble came with darker tones, especially when it came to women.

The Cause

Apparently, companies working with facial recognition software are all suffering from the same issue: a lack of images featuring darker people of color. At least this is what Microsoft is citing as the problem in its own AI. Microsoft set about correcting this using new data to draw upon featuring more dark skin tones.

The new results look promising in both areas. With tests on images of men and women with darker skin seeing error drop up to 20 times. Across the board for all women, the error was dropped by nine times.

While the applications of Face API are numerous, there is still that fear that it could be misused by law enforcement to target Black and brown people. It’s a problem Amazon experienced in Orlando and among its employees and the same thing hit Microsoft recently.

The technology might become precision point in recognizing darker skin tones but it has a long way to go to get over the trust hurdle. One that isn’t of Microsoft’s making.

Starting with Kabir News in 2013, James has focused on tech, gaming, and entertainment. When not writing, he enjoys catching up on sci-fi and horror shows and comics. He can be followed on Twitter @MetalSwift.

Leave a Comment