“My robot couldn’t see me unless I wore a white mask.”

The hours that Joy Buolamwini had spent writing, testing, and coding her “Aspire Mirror” were countless, and her creation had betrayed her. It wasn’t just that it didn’t work. It was that this supposedly neutral machine reinforced a social subtext black women encounter all too often: YOU ARE INVISIBLE.

Her program was supposed to utilize facial recognition software to replace a user’s face with that of an animal that inspires them. The problem she identified was simple. Why it mattered and what to do about it were not. Because there was a lack of diversity in tech, there was a lack of diversity in digital tools.

Joy pushed through her projects substituting her own face for her mask and that of a white roommate before moving on to other PhD research for the MIT Media Lab. But when she attended an artificial intelligence demo years later in Hong Kong, and ran into the same problem, she realized that it was one with far-reaching consequences.

Right now, 16 states allow the FBI to access DMV photos with facial recognition technology in what Georgetown Law refers to as “The Perpetual Lineup.” Joy’s research found that while facial recognition almost always correctly identified a white man’s race and gender (94% at worst), as skin tones got darker, recognition of race and gender became less accurate, with black women being the least identifiable at only 64%.

What happens if the same faulty code misidentifies people of color already facing bias in our justice and immigration systems? How much identification and accessibility software is rendered null & void for anyone without “normal” face specifications? Who is establishing standards for these technologies that might not only be using flawed historical data or personal biases to code, but reinforcing those biases with their outputs? These weren’t just hypothetical questions.

2 years ago, when Microsoft launched an artificial intelligence on Twitter designed to learn from users, then generate its own tweets, it was pulled within 16 hours because what it had learned was racism & sexism. Even Google Images was caught under fire in 2015 when the algorithm they used to “learn” from aggregated searches tagged photographs of black people as gorillas. The next year, an Asian man’s New Zealand passport application photo was rejected because biased facial recognition incorrectly determined his eyes were closed.

So Joy stepped up. She presented her “Gender Shades” research to the major facial recognition software developers Microsoft, IBM, and Chinese company Face++. Only the IBM Watson team responded, validating Joy’s research and integrating it into tests of their newest facial recognition technology.

But she didn’t stop there. Joy’s since created the Algorithmic Justice League, allying major innovators and organizations in artificial intelligence and facial recognition technology under a code of ethics to ensure fairness, accessibility, transparency, and accountability in an increasingly digital age.

“Can machines ever see my queens as I view them? / Can machines ever see our grandmothers as we knew them?” she pleads. And no one is working harder to answer those questions with a resounding yes so that we ALL have a place in the future.

---

KEEP GOING BLACK IN HISTORY: