Last weekend a system of algorithms showed that, at the moment snippets and previewed them, suspiciously, favored white men over others.
Robert Diaz. Peninsula 360 Press.
A big scandal happened this weekend: users of the San Francisco-based social network Twitter, founded in 2006 by Jack Dorsey, Noah Glass, Biz Stone and Eva Willians, debated an algorithm that allows previews to give the user an image of the tweet, whether it's a note or a blog or page.
It all started with the preview images of another platform that, after the COVID-19 pandemic, took on a major importance in the Zoom world, as one of the users, Colin Madland, realized that when he shared the image where he was with his black-skinned boss, only his whiter skin came out.
Colin, made several attempts on Twitter to check that it was random and not really a prefigured schedule. He changed the image and uploaded variations of it to find that in every case the preview removed the figure of his boss.
Tests continued to be done. Researcher Vinay Prabuh tested 92 pairs of images and found that 52 of them selected the black face. That is to say, that this problem is not such. Despite this, the debates multiplied on the Social Network and different developers of the company came out to say that Twitter had already gone through this dilemma, since in the past it used facial recognition and Artificial Intelligence to obtain the images of the previews.
Since 2018 they stopped doing so because of flaws in the AI and the biases it could generate, intentionally or not. The algorithm Twitter now uses to create its previews is trained with data from eye-tracking systems, and responds to areas with higher contrast, faces, animals or text, but does not actively search for these parameters itself.
