After accepting that it found a gender and racial bias in how pictures are cropped on its platform, Twitter has acted swiftly and has now introduced a new bounty program that will reward DEF CON AI Village participants who discover and disclose signs of bias in its image-cropping algorithm.
In May this year, the company concluded the algorithm was biased after testing on randomly linked images of people of various races and genders. The programme favored white people over black people and women over men.
In a recent blog post, Rumman Chowdhury, a software engineering director for Twitter’s machine learning ethics, transparency and accountability team, wrote: “Finding bias in machine learning (ML) models is difficult, and sometimes, companies find out about unintended ethical harms once they’ve already reached the public. We want to change that. As part of this year’s DEF CON AI Village, we’re trying something radical by introducing the industry’s first algorithmic bias bounty competition.”
For the challenge, Twitter said it is re-sharing its saliency model and the code used to generate a crop of an image given a predicted maximally salient point and asking participants to build their assessment.
The winning teams will receive cash prizes via HackerOne – USD3,500 for the winner, USD1,000 for second place, USD500 for third place and USD1,000 each for Most Innovative and Most Generalisable.
“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,” wrote Chowdhury.
“We want to take this work a step further by inviting and incentivising the community to help identify potential harms of this algorithm beyond what we identified ourselves.
“We’re inspired by how the research and hacker communities helped the security field establish best practices for identifying and mitigating vulnerabilities in order to protect the public. We want to cultivate a similar community, focused on ML ethics, to help us identify a broader range of issues than we would be able to on our own. With this challenge, we aim to set a precedent at Twitter, and in the industry, for proactive and collective identification of algorithmic harms.
“For this challenge, we are re-sharing our saliency model and the code used to generate a crop of an image given a predicted maximally salient point and asking participants to build their own assessment. Successful entries will consider both quantitative and qualitative methods in their approach.”
For more details on the challenge, including how to enter and the rubric Twitter will use to score entries, visit the submission page on HackerOne. The challenge will be open for entries from July 30, to August 6, 2021, 11:59pm PT (10:59am on August 7, Dubai time).
All participants must enroll with HackerOne to make a valid submission.
