Twitter in a new attempt to find the algorithmic bias in image-cropping, has proposed a plan to conduct a bug bounty competition. The winners will be rewarded by the company. Twitter usually pays researchers who bring out security issues in its system, and this time they are looking for people to search for unidentified bias in its image-cropping algorithm
Twitter has earlier said that it would look for potential “unintentional harms” on its algorithms. They took image-cropping as the first among them. The image-cropping algorithm was first used in 2018 which helped to focus on certain parts of images in previews. They continued using it since then but there were some criticisms against it saying that it focuses more on lighter-skinned people in photos.
The company in a recent blogpost had mentioned that “In May, we shared our approach to identifying bias in our saliency algorithm (also known as our image cropping algorithm), and we made our code available for others to reproduce our work”. They wanted to bring the community to help them in identifying such harms in their algorithm yet to be identified.
The company also claims that this will be the industry’s first-of-its-kind competition and is offering cash prizes worth $3,500. Rumman Chowdhary who is the director of Twitter’s Machine Learning Ethics, Transparency, and Accountability team said that they are conducting this contest because they think that the people who identify such glitches should be aptly rewarded. He added that the company alone can’t solve these challenges. The winners of the competition will be declared on August 8th at the DEF CON AI Village workshop hosted by Twitter.
