UNMASKING AI

A computer scientist chronicles her journey from eager graduate student to crusader against algorithmic bias.

The daughter of Ghanaian immigrants to the U.S., Buolamwini, the founder of the Algorithmic Justice League, first caught the AI bug as a 9-year-old watching a PBS program about a “social robot” named Kismet, “a dazzling and intricate web of metal and wires topped off with enchanting eyes.” At 25, she was a graduate student in MIT’s Media Lab. At the end of her first semester, Buolamwini made a troubling discovery. Working on a final class project that involved face-tracking software, she discovered that the software was unable to “see” her “dark-skinned” face. However, it could see her face when she donned a white Halloween mask, which meant she could only finish coding the project in “whiteface.” At first, Buolamwini was reluctant to embrace the political dimension of her work, despite more than one encounter in which she realized that the software libraries she was working with “were not optimized for people like me with darker skin.” Eventually, she decided to fully focus on bias in technology, specifically on “AI systems applied to human faces.” At the time, facial recognition systems were being marketed to law enforcement agencies in the U.S. and U.K. Huge tech companies like Amazon and IBM were entering the field, touting programs that could analyze faces and guess demographics like age and race. However, as the author discovered in her groundbreaking research, the systems had largely been trained on data sets predominantly made up of white men; unsurprisingly, they were best at analyzing the faces of white men. Buolamwini is clearly an exceptional scientist and passionate champion of the cause; if her prose is not always inspiring, the content certainly is.

A timely call to action about the near-and-present dangers of AI systems.