Chhavideep SinghTwitter, the micro-blogging social networking website is yet again clarifying the accuracy and non-bias of one of its features. Users never let any platform off the hook, really! And why should it? People are the ultimate beta test for any platform. Well, Twitter’s auto image cropping feature turns out to reflect a bias towards women and white people, the case was reported back in October 2020 and the social networking platform promised to run some tests and look into the matter, and it recently did.
On Wednesday, the company re-evaluated the auto-image cropping algorithm and released the results out in public. As mentioned in multiple reports, Twitter’s initial investigation on the matter did not show relevant results and there was no bias detected whatsoever, while the recent report shows the exact opposite results and it turns out that Twitter’s auto image cropping algorithm was actually flawed and reflecting a bias towards women and white people in general.
According to multiple reports, when testing the algorithm for bias, the company found out that the system was actually treating women and white people differently and there was a clear difference in preference. Twitter’s report revealed that they found a 4% preference towards White people and about 8% difference in preference towards women.
Users filed their complaint about bias towards white people and the flaw here extends those claims. Twitter’s test report also revealed that the auto-cropping algorithm cropped only three out of 100 pictures to a location other than the face of the person. However, the company clarified that the algorithm did not crop any image to specific body parts but to important elements like numbers on sports jerseys, as mentioned in a report by Engadget.
This is a new problem faced by the social networking platform to which they blamed the game on Machine Learning technology. Not really, Twitter’s reason for this auto-generated bias and cropping choices could be high-contrast images, but the company takes responsibility for its actions.
At the end of its statement, Twitter mentioned that one of its researcher’s conclusions is that not everything on the platform is a good candidate for an algorithm. Thus, decisions like cropping an image must be a decision taken by the users according to their own preference. Twitter quotes it as a decision best made by people.
Furthermore, Twitter has recently rolled out full image previews on Android and iOS platforms and the company promises to implement some additional changes to its algorithm and make media better in the future.