Microsoft says its racist facial recognition technology is now less racist – BGR

Earlier this year, Microsoft launched a new facial recognition software to promising developers, but that turned out to be racist. After the "Tay" debacle of the company in 2016 – you probably remember the infallible Twitter robot from Microsoft that has turned into a mad Nazi in less than 24 hours – that 's it! was the last thing that Microsoft needed. The company's "Face API" offered developers access to what the company claimed was advanced facial recognition technology that could be integrated into their own products. But researchers at MIT and Microsoft's New York research lab have discovered that the company's new facial recognition technology has serious race issues. And the kind.

According to an article published by the researchers, Face Face's Microsoft Face Recognition technology has been remarkably competent at identifying white males. In fact, the team discovered that Microsoft technology could identify white male faces with a 0% error rate. This is fantastic, but the result was overshadowed by the fact that the error rate soared as high as 20.8% when technology tried to identify black women. Microsoft clearly had work to do to get it right, and in a blog post on Tuesday night, the company claims to have made significant improvements to its facial recognition technology.

"Microsoft announced Tuesday that it has updated its facial recognition technology with significant improvements in the system's ability to recognize gender through skin tones," John Roach wrote in an article on the company's blog. . This improvement addresses the recent concerns that facial recognition technologies available on the market more specifically recognize the gender of people with lighter skin tones than darker tones, and that they are more effective in males with lighter skin.

He continued: "With the new improvements, Microsoft said it was able to reduce error rates for men and women with darker skin up to 20 times. women, the company said that the error rates have been reduced by nine times.At the whole, the company said that with these improvements, it has been able to significantly reduce the differences in accuracy between demographic data.

Translation: Face API is now less racist.

Microsoft says the high error rates of its technology on female faces with darker skin is indicative of a serious challenge facing the industry as a whole. Artificial intelligence technology needs a data set as wide as possible in order to be well trained. Of course, the responsibility for this problem still lies with Microsoft: the company did not provide enough image data on black people to create a useful facial recognition product.

"The Face API team made three major changes," wrote Roach in his article. "They have expanded and revised the training and benchmark databases, launched new data collection efforts to improve workout data by focusing specifically on complexion, gender and age, and improved the classifier for more accurate results.

Updated Face API software from the Azure Cognitive Services team is available immediately on the Microsoft site.