Race After Technology: Abolitionist Tools for the New Jim Code

In August of 2016, Beauty AI received the results of its international beauty contest judged by artificial intelligence. They found that the AI overwhelmingly favored white people. This bias—in part a result of the beauty samples the creators provided—has broader implications. Similar programs are introduced into important areas such as the healthcare or prison industries. Machine-learning systems and other such technologies at private companies are often outsourced for processes that should be carried out under human supervision with checks and balances.

For centuries, humans have been fascinated with robots. Robots raise questions about dehumanization. They also offer analogies for racism and racialized, historical master/slave relations. If robots are a product of the society that creates them, they can easily be racist, too. They must be trained out of such biases. For robots to be racist does not mean they have “explicit intent to harm” but rather they can be programmed to perform discriminatory tasks. The more they are designed to be like humans, the more they will perpetuate their creators’ biases. 

Benjamin proposes a “race-conscious approach” to designing AI that does not settle for color-blindness (35). She explains how default settings and automated processes can be mistaken for objectivity but can also serve as a mirror for society to see—as concrete data—its own prejudices. Often, when something is better for someone it is worse for someone else, as with segregated fountains. Likewise, when technology is improved for some, it often correlates to a worse experience for others—such as automatic soap dispensers that do not recognize Black people’s hands. Such oversights are important; they help us better understand why technology is designed the way it is.

People have developed social credit systems in real life—such as in China—that track people’s behavior and allow them to rate each other. These credit systems judge who is worthy of what privileges and rely on a biased rubric created by biased actors. Yet these subjective ratings are seen as truly representative of a person’s value.