Twitter will offer cash prizes to hackers who can find image cropping algorithm biases

Twitter has announced that they will be holding their first “algorithmic bias bounty competition”. They hope to incentivise hackers to help identify bias in their image cropping algorithm by handing out cash prizes to the winning teams.

“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,” wrote Twitter in their blog.

Twitter has previously been accused of having an image cropping algorithm with gender and racial bias. The social media platform even reported findings from their own study in May of this year confirming the accusations.

The discoveries include that there was an 8% difference in demographic parity in favour of women. And in the comparison of black and white individuals, the difference is 4% in favour of white people.

There have also been several examples of the Twitter crop bias seen in tweets that show an obvious preference for lighter skin tones. This even included skin tones on fictional characters.

To mitigate the issue at first, Twitter has decided to ditch its auto-cropping feature. They also now have a feature that allows users to see a true preview of the image so that they will know how their Tweets will appear before they post. 

“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,” explained Twitter.

For the challenge, Twitter is re-sharing their saliency model and the code used to generate a crop of an image given a predicted maximally salient point. Those interested in the challenge would need to demonstrate what potential harms an algorithm may introduce.

The winning teams will receive cash prizes and will be able to present their work during the workshop at DEF CON—but they don’t have to attend if they don’t want to. The winning team will be awarded USD 3,500 (RM14,784), and there are separate USD 1,000 (RM4,224) prizes for the most innovative and most generalizable findings.

Opening up a competition will give Twitter a broader range of perspectives. But the cash prize amount has caused a bit of a stir online, as netizens have mentioned that Twitter should “add a few zeros to the prizes to not make it look like an empty marketing stunt”.

If you’re interested in joining, you can visit the competition’s HackerOne page to see the full list of rules and added information. Submissions are open until 7 August 2021 at 4:59PM Malaysian time.

[ SOURCE, 2 ]