Twitter has recently discovered that its image cropping algorithm has a major flaw. Several users had repeatedly pointed out that the algorithm that has been used by the company is biased towards certain people and genders. Twitter had initially denied seeing any spotting any sign of racial or gender bias but recently released the results of the study that tells what exactly is wrong with Twitter's AI tool.
The study reveals that Twitter's saliency algorithm which is used to crop images chose white individuals over Black individuals in images and male-presenting images over female-presenting images. Objectification biases were also discovered in the study. It was reported by people on Twitter that the image cropping tool chose a woman's chest or legs as a salient feature.
Twitter uses the saliency algorithm since 2018 to crop images. In its study, it found out that there was an 8 per cent difference from demographic parity in favour of women as compared to men. The algorithm favoured white people over the back and a 4 per cent difference from demographic parity in favor of white individuals was seen. Similarly, the algo gave preference to white women over men.
Twitter also tested the "male gaze" which was accused of cropping images of men or women on areas other than their faces. But Twitter said that for every 100 images per group, about three cropped at a location other than the head.
The company said that its algorithm preferred high-contrast images so this could be one of the potential reasons behind the biases, but that does not seem very believable. To counter the problem, Twitter had rolled out full-size image previews on iOS and Android.
"The goal of this was to give people more control over how their images appear while also improving the experience of people seeing the images in their timeline. After getting positive feedback on this experience, we launched this feature to everyone. This update also includes a true preview of the image in the Tweet composer field, so Tweet authors know how their Tweets will look before they publish. This release reduces our dependency on ML for a function that we agree is best performed by people using our products. We're working on further improvements to media on Twitter that builds on this initial effort, and we hope to roll it out to everyone soon," Twitter said in its blog.
Twitter also admitted that not everything on Twitter is a good candidate for an algorithm and people make the best decision about how to crop an image.
Copyright©2021 Living Media India Limited. For reprint rights: Syndications Today