Twitter CEO Dantley Davis has his excuses offered for a photo algorithm on Twitter that might systematically emphasize white people in images of people of different skin tones.
The issue came to light when a Twitter user posted as a test image last week that featured both former US President Barack Obama and white US Senator Mitch McConnell. The Twitter algorithm cut out the preview image on McConnell’s face, not Obama’s.
A preview image is a smaller version of a posted photo that has been cropped at a certain point. If someone clicks on the preview, they see the full photo.
Various adjustments to the image did not lead to different results. Other Twitter users tried the experiment on it with different photos and saw the same thing happen.
Davis and his team are now investigating the algorithm, among other things, to rule out that the problem is not due to the contrast in the images, such as the presence of a beard on a white face. “Whatever the reason, this is a problem,” he writes on Twitter.
The Twitter CEO would consider simply stopping using these types of cropped preview images.
Algorithms that use facial recognition are more likely to go wrong with black faces, partly because they are not trained enough on black people. According to Davis, that’s not the problem here, because Twitter doesn’t use facial recognition for photo previews.