Twitter has called for computer experts and hackers to help it tackle bias in its image-cropping algorithm.

As part of the new challenge, the social media company released the code that decides how to crop images publicly. It asked participants to find out how the algorithm could harm people through gender or racial stereotyping.

In return, the company is offering bounties ranging between $500 and $3500 to anyone who finds evidence of harm. In addition, the people will be invited to DEF CON, a major hacker convention taking place in Las Vegas on August 8th, to present their work.

Additional prizes will be offered to the people behind the most innovative and most generalisable solutions.

The new challenge is the first in the Twitter’s Responsible ML initiative, which the social media company previously announced in April this year.

Led by Twitter’s ML Ethics, Transparency and Accountability (META) team, the project aims to assess whether the company’s algorithms have unintentional harmful consequences.

The company estimated that its algorithms affect hundreds of millions of Tweets every day – as such, the potential for bias or discrimination is huge.

“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,” a Twitter statement from Director of Twitter META Rumman Chowdhury and Twitter META Product Manager Jutta Williams read.

“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.”

Recommended

• What are the top-selling games consoles of all time?
• UK spy agencies under fire for “unlawfully” collecting private data
• UK space command launches to protect Britain’s strategic interests

One commonly cited example is racial discrimination over the company’s image-cropping algorithm. When an image of a person is attached to a tweet, Twitter’s AI will crop the image to focus on the face.

However, it was discovered that the algorithm favoured white people over other ethnicities. When an image featuring major US political figures Mitch McConnell and Barack Obama was added to a Tweet, it exclusively focused on white McConnell’s face over Obama’s.

This was confirmed by a Twitter study in May, which found a 4% bias in favour of white people. In addition, the company found its algorithms showed an 8% bias in favour of women over men.

This prompted the company to stop cropping images and displaying them at their standard aspect ratio.