Inside something blog post On Wednesday, Rumman Chowdhury, software design manager for Twitter’s machine learning ethics, transparency and accountability group, wrote that the company felt the algorithm was biased after testing it for groups related to gender and race. Publish and attached research document In detail, the way the cropping system was tested with randomly linked images of people of different races and genders favored whites over, say, blacks and women over men.-
Twitter’s conclusion comes months after the company said he is investigating the algorithm, and it is another example of how biases can be hidden in computer systems designed to perform tasks that people often do exceptionally well. –
“We considered the tradeoffs between the speed and consistency of automatic cultivation and the potential risks that we saw in this study,” Chowdhury wrote. “One of our conclusions is that not everyone on Twitter is a good candidate for the algorithm, and in this case, people’s best decision is to crop the image.”
In March, Twitter began testing a new way to display the full image, rather than an automatically cropped preview, on mobile devices when a user tweets a single image. The company said that after the positive feedback introduced the feature to all its iOS and Android users in May. (However, it centers very long or wide images.) –
A Twitter spokesperson told CNN Business on Wednesday that the change came first on the Twitter mobile app because most people hide and view images.