Researchers at the Georgia Institute of Technology found that state-of-the-art object recognition systems are less accurate at detecting pedestrians with darker skin tones.
Crash-testing: The researchers tested eight image-recognition systems (each trained on a standard data set) against a large pool of pedestrian images. They divided the pedestrians into two groups for lighter and darker skin tones according to the Fitzpatrick skin type scale, a way of classifying human skin color.
Color coded: The detection accuracy of the systems was found to be lower by an average of five percentage points for the group with darker skin. This held true even when controlling for time of day and obstructed view.
Under the hood: Through further analysis, the researchers determined that the bias was probably caused by two things: too few examples of dark-skinned pedetrians and too little emphasis on learning from those examples. They say the bias could be mitigated by adjusting both the data and the algorithm.
An earlier version of this story originally appeared in our AI newsletter The Algorithm. To have it directly delivered to your inbox, sign up here for free.
Why Meta’s latest large language model survived only three days online
Galactica was supposed to help scientists. Instead, it mindlessly spat out biased and incorrect nonsense.
DeepMind’s game-playing AI has beaten a 50-year-old record in computer science
The new version of AlphaZero discovered a faster way to do matrix multiplication, a core problem in computing that affects thousands of everyday computer tasks.
A bot that watched 70,000 hours of Minecraft could unlock AI’s next big thing
Online videos are a vast and untapped source of training data—and OpenAI says it has a new way to use it.
The White House just unveiled a new AI Bill of Rights
It's the first big step to hold AI to account.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.