The times we live in today are very different from the 60s and 70s. A lot has changed since then but one thing that might not have changed is distinguishing people on the basis of their color and gender. In the past, people might have been the culprit but in today’s age software are.
As identity verification systems become a lot common, Amazon has emerged as a frontrunner within the field, serving customers around the US, including police departments and Immigration and Customs social control (ICE). Amazon’s facial identification system called Rekognition is however being accused of being racist and misogynist due to a research conducted by Buolamwini. And it might be enough for people who believe in race equality to be concerning. The latest study published this week by the Massachusetts Institute of Technology Media workplace, states that Rekognition performed worse for dark skinned and white skinned females. To be exact it recognized 19 percent of females as males and mistook 31 percent dark skinned females as males. But it made no mistakes what so ever when it comes to recognizing gender of the lighter skinned men.
Amazon is shying away from addressing this issue and has stated that the research about their facial recognition system doesn’t infer anything about its accuracy of technology. It further noted that research didn’t test the latest Rekognition software. Amazon is claiming that the researchers used their outdated facial recognition software for researching. Amazon has also stated that Police department uses a different facial recognition system to compare people to a mugshot. And that the gender identification test was rather facial analysis (which focuses characteristics like expressions and facial hair), and not facial identification (that finds similarities between scanned faces to mugshots).
A similar kind of test was conducted previous year by ACLU. And it found that Rekognition wrongly matched 28 congress members with police mugshots. Amazon blamed it on poorly calibrated algorithm.
These are two separated software packages, says Amazon. “It’s not possible to draw a conclusion on the accuracy of facial recognition for any use case — including law enforcement — based on results obtained using facial analysis,” Matt Wood, general manager of deep learning and AI at Amazon Web Services, said in a press statement.
However, pundits say Amazon has not done anything significant to make its already biased facial recognition system unbiased towards dark skin individuals. Also, Amazon has not addressed the actual problem presented by researchers.
But it’s not only Amazon that is being scrutinized last February Buolamwin conducted a research on Microsoft, IBM, and Chinese firm Megvil facial recognition system and, upon its analysis, identifies the same racial and gender biases found in the Amazon Rekcognition. The analysis was done last February. Since then Microsoft and IBM both have come forward to address this issue and said that Microsoft and IBM would improve their facial recognition software. According to this study both Microsoft and IBM have improved their facial recognition software. The lesson we learn here is that both Microsoft and IBM understood the criticism and got better. Where as Amazon is still not accepting the fact that their software might have problems that need to be rectified.
Since February, various tech firms have voiced their concerns regarding facial recognition. As bias in the software is usually the results of biased training data, IBM published a curated dataset that aforementioned would boost accuracy. Microsoft too is making their efforts count by recognizing that the industry needs regulation of the technology, to make sure the industry standards don’t take a nose dive to the bottom.
To remind the researchers and experts the magnitude of this issue Buolamwini and co-author Inioluwa Deborah Raji write in their recent paper, “just because a facial recognition system performs equally well on different skin colors, that doesn’t stop it from being a tool of injustice or suppression.”
Further they write: “The potential for weaponization and abuse of facial analysis technologies cannot be ignored nor the threats to privacy or breaches of civil liberties diminished even as accuracy disparities decrease.”
