Twitter is set to largely abandon its image cropping algorithm after its own research found that the system is biased in favor of presenting white people over black people. The algorithm also demonstrated favor towards women over men.
It has been a running joke on Twitter that the site’s image cropping algorithm favored white people over black people, now it appears that Twitter has confirmed that the algorithm was doing exactly that. The Times reports that Twitter’s image cropping algorithm, which has been used by Twitter since 2018 to crop the images that people see when scrolling through timeline, does indeed appear to favor white people in photographs.
The AI tool worked by estimating what a person might want to see first within a picture, having been trained on how the human eye looks at photographs. It would then crop the preview to focus on that section. However, many users complained that the tool seemed to prefer white people, cropping out black people over white people regardless of how the picture was presented. The same research showed that the algorithm showed bias towards women over men.
Twitter claimed last year to have tested its cropping tool for bias before releasing it to the public and had not found any evidence that the tool preferred white people over black people in photographs. However, in a blog post this week, Twitter states that it has concluded that the algorithm was biased.
Rumman Chowdhury, a software engineering director for Twitter’s machine learning ethics, transparency and accountability team, stated that tests showed that the cropping system had a four percent difference in favor of white people. The research showed an eight percent difference in favor of women over men.
The post stated that one reason for this might be that the AI told tends to favor images with high contrast, which is more strongly present in people with lighter skin on dark backgrounds or people with darker eyes on lighter skin.
“We considered the tradeoffs between the speed and consistency of automated cropping with the potential risks we saw in this research,” Chowdhury said. “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.”
Read the full blog post here.
Lucas Nolan is a reporter for Breitbart News covering issues of free speech and online censorship. Follow him on Twitter @LucasNolan or contact via secure email at the address firstname.lastname@example.org