Microblogging website Twitter said on Friday that Twitter integrates competition in its image-cropping algorithm to identify biases. Twitter has revealed plans to compete for prizes with a slightly different bug. Instead of paying researchers to reveal security problems, this time Twitter will be giving awards those who find undiscovered examples of bias in its image cropping algorithms.
Back in April, Twitter announced that it would investigate potential “unwanted damage” caused by its algorithm, starting with cropping the image. He started using the algorithm in 2018 to focus on the most interesting parts of the image in the visualization.
Some users have criticized the way Twitter handles auto-cropping, arguing that the algorithm focuses on people with lighter photos.
Twitter Inc said Friday it would launch a competition for computer researchers and hackers to identify biases in the algorithms used to crop images after researchers previously found algorithms that tended to exclude blacks and men.
Competitors are part of a broader effort across the tech business to ensure that the applied science of synthetic intelligence works ethically.
The social media company said on its web blog that competitor pricing is aimed at “understanding the potential harm this algorithm can do beyond what we’ve seen.”
After criticizing image visualization in a non-black publication last year, the company announced in May that a study by three of its machine learning researchers found an 8% difference in demographic parity in favor of women and 4% in favor of white.
Twitter released the computer code that decides how images are cropped on its Twitter feed, saying participants were asked on Friday to find out how the algorithm could harm, such as:
Winners will receive cash prizes of between $500 and $3,500 and will be invited to showcase their work at a Twitter-hosted seminar at DEF CON in August, one of the largest hacking conferences held annually in Las Vegas.