(NEW YORK) — After social media exploded with allegations that Twitter’s image-cropping algorithm had a racial bias, the company said it investigated and ultimately decided to drop automated cropping.
Late last year, posts on Twitter that included large photos with a white and Black person went viral as users pointed out that the algorithm tended only to show a preview of the white person when it had to crop the photo to fit the platform’s aspect ratio for images.
A Twitter spokesperson said in a tweet in September 2020 that the feedback shows “it’s clear that we’ve got more analysis to do.”
The social media giant on Wednesday released the outcomes of its quantitative research into the algorithm’s potential bias, which found it had a 4% difference in favor of white individuals in comparisons of Black and white individuals. Moreover, there was a 7% difference in favor of white women in comparisons with Black women and a 2% difference in favor of white men in comparisons with Black men.
In comparisons of men and women, the researchers found there was an 8% difference in favor of women.
The so-called saliency algorithm works by estimating what a person might want to see first within a picture so that it can be cropped to an easily-viewable size, according to Twitter. The algorithm was trained on human eye-tracking data and scores all regions of an image and then chooses the point with the highest score as the center of the crop.
As a result of the research, the company said it ultimately decided photo-cropping on Twitter is a task best done by humans.
“We considered the tradeoffs between the speed and consistency of automated cropping with the potential risks we saw in this research,” Rumman Chowdhury, the company’s director of software engineering, said in a blog post. “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.”
Chowdhury said they began testing a new way to display standard aspect ratio photos in full on iOS and Android in March, without the saliency algorithm crop, and eventually decided to roll it out to everyone.
“This update also includes a true preview of the image in the Tweet composer field, so Tweet authors know how their Tweets will look before they publish,” Chowdhury said. “This release reduces our dependency on ML [programming language] for a function that we agree is best performed by people using our products.”
The findings come as the tech sector grapples with racial biases creeping into work done by computers.
Racial bias, even in computer algorithms, can have real-world implications for people of color. A study published in the peer-reviewed journal Science in 2019 found that an algorithm widely used in U.S. hospitals to allocate health care to patients systemically discriminated against Black people.
Copyright © 2021, ABC Audio. All rights reserved.