Twitter Drops AI After Discovering Racial, Gender Biases

Popular social media platform Twitter confirmed Wednesday, May 19, 2021, via a blog post that its artificial intelligence algorithm has been biased against certain genders and races. In particular, the said system reportedly took to favoring White people and women over Black people and men.

The software engineering director for the app’s machine learning ethics, transparency, and accountability team Rumman Chowdhury took to admitting the racist and biased algorithm in a blog post.

According to CNN, the admission comes months after the Twitter team initially discovered and looked into the said algorithm. The news site revealed that the developers tested the system on images of people of various races and genders, with the artificial intelligence algorithm showing a preference for White people over Black people, as well as women over men.

Twitter Drops AI After Discovering Biases

In a statement, Chowdhury said, “We considered the tradeoffs between the speed and consistency of automated cropping with the potential risks we saw in this research. 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.”

The Twitter blog post was accompanied by a research paper from the company’s machine learning researchers after having their system criticized last year. The paper is titled, “Image Cropping on Twitter: Fairness Metrics, their Limitations, and the Importance of Representation, Design, and Agency.”

In an article by the Washington Post, the new site said that 8% of the tested images favored women, while 4% of these were inclined towards White people. Within this group, the researchers found that the algorithm preferred White women over Black women by as much as 7%. Meanwhile, the system was also biased towards White men over Black males by 2%.

Researchers say that this could be attributed to the image backgrounds, as well as the different eye colors of the people in the images. However, Reuters said the researchers acknowledged that these are not excuses.

A statement in the paper reads, “Machine learning based cropping is fundamentally flawed because it removes user agency and restricts user’s expression of their own identity and values, instead imposing a normative gaze about which part of the image is considered the most interesting.”

In attempts to address the issue, Twitter has since removed the infamous Twitter crop. Rather, it has taken to showing photos in full ratio, reports Reuters. The new system was initially rolled out in March, with the fill feature made available to iOS and Android users just in May this year, notes CNN.