Nintendo-Power

Daily Gaming news, videos, reviews, tips & guides. Let's share our love of BigN games!

Photo crop algorithm removed due to racial bias

Photo crop algorithm removed due to racial bias

Twitter app on smartphone. – Matt Rourke / AP / SIPA

On top of that Social Websites, Technology is not always good when fully automated. Twitter has noticed it
Artificial intelligence Doesn’t always sound like ethics. The company has really established that its method of cultivating photos is somewhat biased in favor of white people and women.
A problem that has already been pointed out by experts for months 20 minutes. So the site wants to do without it.

“The algorithm does not apply to everything in it Twitter “, On Wednesday recognized the social network in one of its blogs. “Users are good at deciding how to format an image. “

An algorithm in service from 2018 onwards

This algorithm, which will be in service in 2018, is designed to crop the images into what they consider to be the most important, in order to reduce the size of the images and minimize the thread of tweets. “People on Twitter have noticed that our program selects whites more than blacks,” says director Ramman Chowdhury. Software. “We tested our system on a large database to determine if there was a problem.” The researchers found that there was a 4% difference in favor of whites in general and 7% support for white women against black women. In general the comparison between men and women revealed an 8% gap in favor of women.

Company in San Francisco The algorithm also searched for so-called “male vision” bias, meaning that a woman would select her chest or legs instead of her face. But only 3% of the photos tested had people cut out of the head, and non-physical elements like the number on a sports jersey. “We have not found any evidence for an objective bias,” says Ramman Chaudhary.

Low use of “machine learning”

The social network tested a new way of displaying images without carving in March, and then expanded the tool for everyone, allowing everyone to see what their tweet would look like before posting. “This reduces our reliability on machine learning for a task that our users can do better,” Ramman Chowdhury said. So artificial intelligence is still in progress.

See also  Facebook apps are in trouble