Twitter is trying to determine why its photo preview feature shows signs of racial bias. Several users found that black faces were more likely to get cropped out of photo previews in comparison to white faces.

 A Flaw in Twitter's Algorithm

When you see an image on Twitter, it's just a preview---you have to click on it to see the entire picture. Twitter uses an algorithm to crop photos, dictating what you'll see in image previews.

In 2018, Twitter announced its smart auto-cropping feature in a post on the Twitter Blog. This feature uses "saliency" to hone in on the most prominent parts of photos, and it then crops photo previews accordingly.

Twitter users discovered that the platform's neural network may have a racial bias. One user posted a picture of US Senator Mitch McConnell alongside former President Barack Obama.

The image preview blatantly crops out Obama's face, even when his face is placed at the top of the photo. When the same user tried inverting the picture's colors, Obama's face finally showed up in the image preview.

Other users began conducting their own experiments. Time and time again, users found that the algorithm seemed to consistently prefer white faces over black faces.

Liz Kelley, a Twitter communications team member, sent out a Tweet in reply to the Twitter users' findings. She noted that while the Twitter team already tested the algorithm for bias, it would have to investigate further.

All of these discoveries stemmed from a user's Tweet about a Zoom call issue. Zoom continuously failed to detect his black friend's face when he tried to use a virtual background. When the user turned to Twitter to post screenshots of the problem, Twitter cropped his black friend out of the image preview.

Taking Bias Out of Twitter

While all of these experiments are informal, they seem to highlight a real issue. Twitter will definitely have to take a closer look at its auto-crop feature, and see if it has an ingrained racial bias.

That said, Twitter's auto-crop algorithm isn't the only Twitter feature that's been criticized. After all, many users also aren't fans of the platform's algorithm that automatically arranges your feed.