Twitter just issued a new challenge that will pays people to actually do some good.

As part of the hacker conference DEF CON beginning Aug. 5, Twitter has announced a challenge to the hacking/coding community to help the company make the algorithm that crops the platform’s photos less (apparently) racist.

For photos attached to tweets that are a larger or different shape than the thumbnail proportions, a Twitter algorithm chooses what portion of the image to show in the tweet’s preview. In Sept. 2020, Twitter users pointed out that this algorithm appeared to more consistently show white faces than black faces when there was an odd-shaped image that contained both. Some described this as an instance of algorithmic bias, which is when racism intentionally or inadvertently gets baked into the decisions a computer makes.

In response, Twitter shared the way the algorithm worked, and said that it would look into the matter. But the algorithmic bias bounty program takes efforts to solve this problem a step further by financially incentivizing people to mitigate it.

Bug bounties are programs companies or other groups have that reward people (often, those outside the organizations) for finding bugs in their technical infrastructure. Bug bounties often focus on finding potential security breaches, and organizations pay bug bounty hunters for alerting them to the issues.

Twitter is taking this concept and applying it to the challenge of algorithmic bias in photo cropping. It will award $3,500 to a first place winner that can identify the cause of this apparent bias, plus $1,000 each for second place, “most innovative,” and “Most Generalizable (i.e., applies to the most types of algorithms),” and $500 for third place.

The challenge is a neat use of the bug bounty model: Rather than apologizing and playing defense for an instance of algorithmic bias, Twitter is saying, we want proactive help. And we’ll pay.

“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,” Twitter’s blog post announcing the challenge reads. “With this challenge we aim to set a precedent at Twitter, and in the industry, for proactive and collective identification of algorithmic harms.”

The bounty program at this time is specific to the challenge of photo cropping. But the blog post announcing the challenge describes it as “Twitter’s first algorithmic bias bounty challenge.” First, and maybe not last? Mashable has reached out to Twitter to learn whether it will launch more algorithmic bias bounties to hunt down bias in machine learning, and will update this when we hear back.

Original Source