Twitter is holding a competition wherein its image cropping algorithm hackers and researchers will identify biases. It will be handing out cash prizes to winning teams. According to the company giving teams access to its image cropping model and code will let them find ways that the algorithm could be harmful.
The description of their findings and a dataset will have to be submitted by those competing to demonstrate the issue that can be run through the algorithm. Based on how much it could affect people, what kind of harm is found, and more, Twitter will then assign points. $3,500 will be awarded to the winning team, and there are separate $1,000 prizes for the most generalizable and most innovative findings. A bit of stir has been caused on Twitter by the amount, with few users saying it should have an extra zero.
The company’s normal bug bounty program will pay$2,940 if a bug is found to perform actions for someone else using cross-site scripting. Finding an OAuth issue that lets one take over someone’s Twitter account would net one $7,700. In May, Twitter has done its research into its image-cropping algorithm. After accusations that its previews crops were racist, it has published a paper investigating how the algorithm was biased. Since then, the company has mostly done away with algorithmically cropping previews, but it’s still used on desktops. From a much broader range of perspectives, the company can get feedback by opening up competition.
Twitter is not just looking for subconscious algorithmic bias. For both intentional and unintentional harms, the rubric has point values. According to the company, unintentional harms are crops that could result from a “well-intentioned” user posting a regular image on the platform. Intentional harms are problematic cropping behaviours that could be exploited by someone posting maliciously designed images.
In its announcement blog, the company said that the competition is separate from its bug bounty program if one submits a report about algorithmic biases to the company outside of the competition. Twitter says the report will be closed and marked as not applicable. If anyone is interested in joining, then they can head over to the competition’s HackerOne Page to see the rules, criteria, and more. Until August 6th at 11:59 PM PT, submissions are open, and on August 9th, the winners of the challenge will be announced at the Def Con AI Village.