Twitter Offers ‘Bug Bounty’ Contest For AI Biases

Twitter has launched a contest to lure developers to help it find biases in the algorithm that automatically crops images on the service, following in the spirit of a common practice in the computer security industry.

Software firms commonly offer “bug bounties” to those who discover security flaws in their code, and Twitter said it is looking to create a similar incentive with prizes of up to $3,500 (£2,500).

The company said in a blog post announcing the scheme that it is looking for ways of identifying “unintended ethical harms” in the algorithm, after a group of researchers previously found it tended to exclude Black people and men.

Algorithmic bias

The algorithm is intended to display previews of the most relevant portion of an image in a Twitter feed.

Technology companies have increasingly looked for ways to ensure the artificial intelligence systems they use behave in ethical ways, amidst concerns over the growing role of AI, algorithms and other automated systems in communications and society.

In May Twitter made the code for its image-cropping algorithm available to outside researchers.

“We want to take this work a step further by inviting and incentivising the community to help identify potential harms of this algorithm beyond what we identified ourselves,” the company said.

It said it was “inspired” by the way researchers and hackers help the security field establish best practices for identifying and mitigating vulnerabilities to protect the public.

“We want to cultivate a similar community, focused on ML ethics,” Twitter said.

The company said it wanted to set a precedent for “proactive and collective identification of algorithmic harms”.

Cash prizes

It is re-sharing its image cropping model and code and is asking participants in the contest to build their own assessments.

The contest will be conducted through the HackerOne platform, with a panel of judges evaluating entries.

Aside from three top prizes of $3,500, $1,000 and $500 the contest will issue $1,000 prizes for the two entries found most innovative and most generally applicable across different types of algorithms.

The contest is currently open and winners are to be announced on 8 August at the Twitter-hosted AI Village workshop, part of the annual DEF CON security conference in Las Vegas.