Microsoft ‘Improves’ Racist Facial Recognition Software For WOC

Microsoft has announced that it has taken measures to improves its lacking facial recognition software. Earlier this year, news broke that the company’s facial recognition software had a 20.8 percent error rate, an error rate that only applied when the software was identifying the gender of people of color. The seemingly racist software had a particular issue when it came to women with darker skin.

Apparently the issue stemmed from the fact that Microsoft did not have enough images of black and brown people in its database—a common issue in the technology industry. According to Microsoft, the fix was to further data collection efforts and expand their datasets. Basically, they found more photos of women of color and added them to their facial recognition database.

Because women wear makeup, more jewelry and a wider variety of hairstyles, I kind of understand why identifying us is a bit more challenging, but I’m still baffled that no one thought to include more skin tones in the original datasets. People of color represent a huge part of the technology market and there’s honestly no excuse for this to still be such an issue. Facial recognition has been around for years now!

Of course, it’s great that Microsoft is finally aware of the issue and making ongoing efforts to address it, but the fact that they even have to— is again— proof that white male privilege is very much still in existence. The whole thing is just further proof of our marginalization as women of color. I mean, who decided to load up the database with white men in the first place? I know there has to be a lot of brown people working for Microsoft. I seriously don’t get it.

In any case, the company says that the improvements they made will reduce error rates for people with dark skin by up to 20 percent and for women in particular by nine times. It’s a start. Hopefully they continue to work on it, so that maybe we can hope for that elusive zero percent error rate that white males enjoy.