Twitter’s auto image cropping has a bias towards white men and women

In October, Twitter promised to re-examine the image of this cropping algorithm after users complained it was biased. On Wednesday, the company released the results of that study. When people first started writing about algorithm problems, it was said on Twitter initial investigation showed no evidence of racial or gender bias. now different pictures were painted in the research.

Comparing the preference of the algorithm between men and women, in addition to its treatment of white and Black individuals, the company found an 8 percent difference in favor of women and 4 percent in favor of whites. What it doesn’t detect is that the algorithm shows a male-looking problem. Tested against images of women, Twitter found that the algorithm only cropped about three photos per 100 in a location other than the face, and when it did, it didn’t focus on specific body parts. However, it is tied to elements such as the number of sports jerseys.

The company cited a potential reason for algorithm choices as it might prefer for images with high contrast but said that wasn’t an excuse. “Machine learning-based cropping is informative because it takes away the user’s agency and restricts the user’s expression of their own identities and values, instead of imposing a normative stance. look at what part of the image is considered most interesting, “according to the company. “One of our conclusions is that not everyone on Twitter is a good candidate for an algorithm, and in this case, how to harvest an image is a decision that most people make. . ”

To this end, the company has just been launched full size image previews on Android and iOS, and said it plans to implement further changes to how the platform handles media in the future.

All products recommended by Engadget are selected by our editorial team, which is independent of our parent company. Some of our stories have accompanying membership links. If you buy something through one of the links, we can get a co -commission.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *