Twitter reported that its image-cropping algorithm does in fact have a bias against Black people and men. On Wednesday (May 19), researchers with the platform released their findings in a study conducted on its image-cropping feature, Reuters reported. The company said in announcing the findings, “how to crop an image is a decision best made by people.”
The study, conducted by three AI researchers at Twitter, found that images with women and white people were favored in image previews. They cited possible explanations for the bias including eye color, image background, though the company said there was no excuse for it.
“Machine learning based cropping is fundamentally flawed because it removes user agency and restricts user’s expression of their own identity and values, instead imposing a normative gaze about which part of the image is considered more interesting,” the researchers wrote in a paper.
To address the bias, Twitter recently rolled out a non-cropped image feature for its mobile app that had many users taking advantage of the newfound ease of posting pictures earlier this month.
The findings of Twitter’s study adds to the growing number of efforts to unpack bias in artificial intelligence technology in our daily lives. Outside of social media, facial recognition technology used by law enforcement falsely identifies Black people than other groups, and has led to the wrongful imprisonment of several people.
Efforts to diversify AI research and STEM fields overall continue to be prioritized among companies and universities, especially HBCUs.
Photo: Getty Images