Twitter’s algorithm for the automatic trim of photos preferred not only women in general, but also favored pictures of people who look Dunner and young. In addition, he disadvantaged people with female hair as well as Arab font against Latin letters. These are the results of an unusual competition that Twitter has proclaimed to identify further hitherto unknown prejudice. He was organized at the weekend as part of the Conference DEF CON and continued a debate that had been released last year. At that time it had become clear that the algorithm black and women people reproducibly treated differently.
Money for search for restrictions from Ki
Based on bug-bounty programs in which there are money for found and reported security in software, Twitter had launched the competition to find prejudices of their own algorithm ‘. The company had made the participants and participants for the corresponding source code for the purposefully searching for favors and disadvantages of the software. Twitter did not just want to find further problematic aspects of his own technique, but also put an example of the whole industry. For the best contributions, prize money has been provided in a high number of $ 3500 US dollars.
Now Bogdan Kulynych has won a student of the Swiss Student Lausanne (EPFL). With manipulated photos of faces, he could prove that the algorithm not only feminine persons against masculines, but also more young people against ages and brighter, soft skin against darker preferred. Other analyzes showed that people with headscarves, those with disabilities and generally old people are considered in a similar manner of the algorithm, reports NBC. Accordingly, all such faces are preferably cut off from the algorithm, for example if more on the photo can be seen. A competition contribution could prove that Latin font is favored against Arabic, Wired is still added.
Twitter had already drawn the algorithm analyzed after the first criticism and blessed that the choice of image clipping among users and users is better lifted. The possible advantages through the use of the technique, for example, as a speed, were the risk of discrimination did not destroy, Twitter had explained. In the apps can be selected since May, as an image should be displayed and a preview is also available. In addition, high-format photos in the timeline have been possible since then. That Twitter had not allowed that, had first guided at the first finds of the discrimination.