Microsoft Facial Recognition Software – According to a Microsoft blog post, its facial recognition tools are getting enhanced at recognizing people with darker skin tones than before. The rates of error have been reduced by almost 20 times for women and men with the darker skin and by nearly nine times for all the women.
The Microsoft said that it has been training its AI tools with some larger and some more diverse datasets which have directly led to the progress.
Hanna Wallach, a senior researcher of Microsoft, in the blog post, said that if the Microsoft is training machine learning systems to the imitate decisions that are made in an unfair society, by using data generated process by that society, then those systems will essentially reproduce its biases.
In February 2018, a report from the MIT’s Media Lab tested that facial recognition systems from the Microsoft, IBM, and China’s Megvii and also found that up to the 35 percent of darker-skinned women had their gender misrecognized by the systems. The report confirmed that what many others had suspected for years but the facial recognition systems might be biased by the limited datasets and some of the other factors like systemic racism. In the year 2015, Google recognized a software engineer’s black friends into a photo as “gorillas,” and had to express regret for the error.
The recent Microsoft’s announcement indicates that a reduction in the ethnic bias within its facial recognition system, if the law enforcement gets their hands on the better facial recognition tool, then it is very unclear and might play out of color and whether that could aggravate already challenging practices for everybody. Microsoft partners with the US Immigration and Customs Enforcement (ICE), and by its facial recognition tool which is offered to the government agents as a resource.
The Microsoft CEO Satya Nadella has clarified in a memo that now the Microsoft is not working with U.S. government on any of the projects related to separating the children from their families at the border, but he did not address that how facial recognition might play a task for benefitting the ICE’s work.