The bounty challenge opened in July after Twitter users showed that the site’s automated cropping tool favored the faces of people with lighter complexions over those with darker complexions. It raised some questions about how the software prioritized skin color and certain factors over others. The challenge sought to find what other bugs and biases the cropping system may have in order to fix the issues. First place went to Bogdan Kulynych, whose submission showed how beauty filters could game the algorithm’s scoring model, which, in turn, amplify traditional beauty standards. The submission showed the algorithm preferred young and slim faces with either a light or warm skin tone. Kulynych won $3,500. Second place went to HALT AI, a tech startup in Toronto, which discovered images of the elderly and the disabled were cropped out of photos. The team was given $2,000 for coming in second. Third place, and $500, went to Roya Pakzad, founder of Taraaz Research, who discovered the algorithm favored cropping Latin scripts over Arabic scripts, which could harm linguistic diversity. The detailed results were presented at DEF CON 29 by Rumman Chowdhury, the director of Twitter’s META team. The META team studies the unintentional problems in algorithms and weeds out any sort of gender and racial bias such systems may have. The data obtained from this contest will be used to alleviate bugs and bias in the cropping algorithm and help ensure a more inclusive environment.
