Twitter apologizes after users notice image-cropping algorithm favours white faces over Black

Big oops:

Twitter has apologized after users called its ‘image-cropping’ algorithm racist for automatically focusing on white faces over Black ones.

Users noticed that when two separate photos, one of a white face and the other of a Black face, were displayed in the post, the algorithm would crop the latter out and only show the former on its mobile version.

PhD student Colin Madland was among the first to point out the issue on Sept. 18, after a Black colleague asked him to help stop Zoom from removing his head while using a virtual background. Madland attempted to post a two-up display of him and his colleague with the head erased and noticed that Twitter automatically cropped his colleague out and focused solely on his face.

“Geez .. any guesses why @Twitter defaulted to show only the right side of the picture on mobile?” he tweeted along with a screenshot.

Entrepreneur Tony Arcieri experimented with the algorithm using a two-up image of Barack Obama and U.S. Senator Mitch McConnell. He discovered that the algorithm would consistently crop out Obama and instead show two images of McConnell.

Several other Twitter users also tested the feature out and noticed that the same thing happened with stock models, different characters from The Simpsons, and golden and black retrievers.

Dantley Davis, Twitter’s chief design officer, replied to Madland’s tweet and suggested his facial hair could be affecting the model “because of the contrast with his skin.”

Davis, who said he experimented with the algorithm after seeing Madland’s tweet, added that once he removed Madland’s facial hair from the photo, the Black colleague’s image showed in the preview.

“Our team did test for racial bias before shipping this model,” he said, but noted that the issue is “100% (Twitter’s) fault.” “Now the next step is fixing it,” he wrote in another tweet.

In a statement, a Twitter spokesperson conceded the company had some further testing to do. “Our team did test for bias before shipping the model and did not find evidence of racial or gender bias in our testing. But it’s clear from these examples that we’ve got more analysis to do. We’ll continue to share what we learn, what actions we take, and will open source our analysis so others can review and replicate,” they said, as quoted by the Guardian.

Source: Twitter apologizes after users notice image-cropping algorithm favours white faces over Black