I got curious about computer science when I was nine years old. I was watching PBS and they were interviewing someone from MIT who had created a social robot named Kismet. It had big ears that moved. It could smile. I didn't know you could do that with machines. So, from when I was little, I had in my mind that I wanted to be a robotics engineer and I was going to MIT.
Eventually, I did reach MIT, but I went to Georgia Tech for my undergraduate degree. I was working to get a robot to play peekaboo because social interactions show some forms of perceived intelligence. It was then that I learned about the code bias: Peekaboo doesn't work when your robot doesn't see you.
[A few years later] at MIT, when I was creating [a robot] that would say, "Hello, beautiful," I was struggling to have it detect my face. I tried drawing a face on my hand and it detected that. I happened to have a white [Halloween] mask in my office. I put it on [and it detected that]. I was literally coding in whiteface. I did another project where you could "paint walls" with your smile. Same issue: Either my face wasn't detected, or when it was I was labeled male.
Cathy O'Neil's book Weapons of Math Destruction talks about the ways technology can work differently on different groups of people or make certain social conditions worse, which made me feel less alone. [At the time], I was a resident tutor at Harvard and the dining hall workers were on strike. I heard people protesting and wondered, Do I just follow this comfortable path that I'm on or might I take the risk and fight for algorithmic justice?
I changed my research focus and started testing different systems that analyze faces. That became my MIT master's work, [a project] called Gender Shades. I collected a data set of members of parliament from three African countries and three European countries — and found AI systems worked better overall on lighter-skinned faces. Across the board, they work the worst on people most like me: darker-skinned women.
Black men have been arrested [after being] linked to false facial recognition matches. Fifty percent of adult faces [can be found] in a recognition network that can be searched by law enforcement. You can be considered in a lineup not because you’ve done anything, but maybe because you updated your driver's license.
Algorithmic injustice certainly impacts marginalized communities, but no one is immune. People have been denied access to their social security benefits. [These systems can be] involved in the screening of job applications. If you're using history to train these systems, the future’s going to look like the sexist past.
It's easy to assume that because it's a computer, it must be neutral. It's a seductive notion. Within the beauty space, when you search [the internet] for "beautiful skin," or "beautiful woman," it's Eurocentric beauty [that's represented]. When we train machines, they become a mirror to our own limitations within society. We have to keep moving toward broader representation.
I worked with Olay on their Decode the Bias campaign [to inspire more women to pursue careers in STEM]. In an audit of their Skin Advisor system, which analyzes people's images to make product recommendations, we did find bias. Then Olay made commitments [to mitigate] those issues. For me, as a dark-skinned Black woman, to be the face of a campaign like that... There is hope.
That's what we're pushing for with [my nonprofit], the Algorithmic Justice League: [questioning bias] as the default way of making these systems. It should be part of the conversation from the beginning — not something that happens after someone gets arrested. — As told to Dianna Mazzone
This story originally appeared in the August 2022 issue of Allure. Learn how to subscribe here.
Read more about women in STEM:
- The Geniuses Behind Some of the Most Innovative New Skin-Care Brands
- Why Millennial Women in Beauty Are Embracing Bitcoin
- What It's Really Like to Be a Cosmetic Chemist
Now watch a doctor's entire daily routine:
