Why Twitter wants ethical hackers to fix its algorithmic biases
The company wants to cultivate a community of ethical AI hackers
Tapping into the community
The initiative is not the first time that Twitter’s sought community support for mitigating algorithmic harms.
In May, the META team shared its research and code on the image cropping algorithm’s biases so that others could investigate the issue.
The cropping algorithm estimates what people want to see first within a picture. This calculation then determines how an image is cropped to an easily viewable size.
The model was trained on human eye-tracking data to predict a saliency score on all regions of a picture. It then chooses the point with the highest score as the center of the crop.
After receiving feedback that the algorithm didn’t serve all people equitably, Twitter analyzed the model for biases. The researchers uncovered underlying issues that favored white individuals over Black people.
“We want to take this work a step further by inviting and incentivizing the community to help identify potential harms of this algorithm beyond what we identified ourselves,” Rumman Chowdhury, the head of Twitter’s META team, told TNW.
In the challenge, participants will get access to Twitter’s saliencymodel and the codeused to generate a crop of an image. Their mission is to demonstrate potential harms that such an algorithm may produce.
Democratizing standards
A key goal of the contest is to develop community-driven standards and best practices for assessing ML models. Notably, Twitter has created a grading rubric that articulates algorithmic harms in a way that didn’t previously exist.
There’s already a large community of ethical AI hackers that Twitter hopes to tap into.Historically, however, they haven’t been incentivized to do this sort of work in the same way as whitehat security hackers.
“In fact, people have been doing this sort of work on their own for years, but haven’t been rewarded or paid for it,” said Chowdhury.
Theintroduction of monetary rewards will add further encouragement.
Ultimately,Chowdhury wants to foster a more inclusive and proactive approach to mitigating algorithmic risks:
Thechallengeis open for entries until 11:59PM PT on August 6. The winners will be announced atthe DEF CON AI Village workshopon August 8. Anyone with a HackerOne account can participate in the competition.
Greetings Humanoids! Did you know we have a newsletter all about AI? You can subscribe to itright here.
Story byThomas Macaulay
Thomas is a senior reporter at TNW. He covers European tech, with a focus on AI, cybersecurity, and government policy.Thomas is a senior reporter at TNW. He covers European tech, with a focus on AI, cybersecurity, and government policy.
Get the TNW newsletter
Get the most important tech news in your inbox each week.
