Twitter announces bug bounty programme to find biases in AI algorithms

“We’re inspired by how the research and hacker communities helped the security field establish best practices for identifying and mitigating vulnerabilities to protect the public,” the company said in a blog post. “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,” it added.

The social media platform will share the code for its saliency model, which is used to generate cropped versions of images on the platform. “Successful entries will consider both quantitative and qualitative methods in their approach,” the post says. It asked community members to submit their entries through vulnerability coordination and bug bounty platform HackerOne.

Twitter will be announcing the winners on 8 August at the DEFCON conference’s AI Village this year. It will invite winners to present their work during the conference, offering $3,500 to the first-place winner. Second and third place winners will get $1,000 and $500, respectively, while there are also $1,000 awards for “most innovative” and “most generalizable” entries. The latter are entries that will apply to most types of AI algorithms.

Tech companies such as Twitter regularly run bug bounty programmes to keep their platforms secure, but a program to find algorithmic bias is rather new. The move is in line with what the company’s chief executive officer, Jack Dorsey, had said during the fourth quarter earnings call in February. Dorsey suggested a marketplace approach to recommendation algorithms, offering users a chance to choose the kind of algorithm that they want to use.

“One of the things we brought up last year, to address some of the problems facing Section 230 (of US’ Communications Decency Act), is giving people more choice about what ranking algorithms they’re using,” he said at the time. “You can imagine a more marketplace approach to algorithms, and that’s something that we can not only host but also participate in,” he added.

Biases and inaccuracies in recommendation algorithms used by platforms such as Facebook, Twitter, and Google are a major part of the focus for upcoming regulations, too. The controversy around Google letting go of AI ethics researcher Timnit Gebru has also raised questions about whether big tech firms are really interested in finding inaccuracies and biases in their algorithms, especially when they interfere with revenues.