clock menu more-arrow no yes

Filed under:

Twitter reveals image-cropping algorithm was biased toward white people and women

The company’s previous cropping system favored white people or Black people and women over men.

Twitter Getty Images

In March, Twitter announced a new format that would allow users to post full photos rather than cropped images. The change was made to replace the platform’s image-cropping algorithm, which the company recently admitted was biased, according to a CNN report.

On Wednesday (May 19), Rumman Chowdhury — a software engineering director for Twitter — wrote a blog post revealing that the social media platform’s previous automated cropping system was partial toward white people and women.

The algorithm initially used face detection to determine how images would be cropped in previews, Twitter said in a 2018 blog post. However, since it was prone to errors, they decided to focus on “saliency,” the most noticeable, important or interesting area of a full image.

The cropping algorithm eventually came into question last September after several critiques from users, including Twitter user @bascule, who tried a “horrible experiment” to show the system’s biases to his followers. In his tweet, he posted two photos of Senate Majority Leader Mitch McConnell and Barack Obama. In one image, Obama was featured at the top; in the other, he was at the bottom. Twitter’s system cropped both images to show only McConnell in the previews.

In response to the criticism, the company issued a statement promising to take another look at the system. “We tested for bias before shipping the model & didn’t find evidence of racial or gender bias in our testing,” they wrote. “But it’s clear that we’ve got more analysis to do. We’ll continue to share what we learn, what actions we take, & will open source it so others can review and replicate.”

After testing the algorithm once more for gender and race-based biases, it was determined that there’s about an eight percent demographic parity favoring women over men. As for its bias between white and Black men, the research shows that there’s a four percent parity.

“We considered the tradeoffs between the speed and consistency of automated cropping with the potential risks we saw in this research,” Chowdhury wrote along with the findings. “One of our conclusions is that not everything on Twitter is a good candidate for an algorithm, and in this case, how to crop an image is a decision best made by people.”

Sign up for the newsletter Join the revolution.

Get REVOLT updates weekly so you don’t miss a thing.