Developers Take on Extremism in AI Challenge
By Tech Correspondent
An innovative new challenge has been launched, enlisting developers worldwide to create advanced machine-learning models aimed at combating extremism. The competition, which runs until November 7, 2024, presents participants with a two-fold task: developing models to identify hateful images and countering them by creating images that can deceive these identification models. The initiative aims to enhance the capabilities of AI in moderating and identifying hate-based content online.
The challenge is spearheaded by Humane Intelligence in partnership with a Nordic counterterrorism group. The competition is designed to mimic real-world scenarios where developers work to build security measures while adversaries develop methods to circumvent them. "The do-gooders make one approach, and then the bad guys make an approach," explained Chowdhury, a leading figure in the project. This process is intended to foster robust machine-learning systems capable of adapting to constantly evolving threats.
Participants will submit their models in two categories. The intermediate challenge focuses on creating a model that can identify hateful images, while the advanced challenge tasks developers with creating a model that can trick the identification model. Winners will receive monetary rewards of $4,000 for the intermediate challenge and $6,000 for the advanced challenge. Their submissions will also undergo evaluation by Revontulet for potential adoption into its suite of anti-extremism tools.
A critical challenge in identifying hate-based propaganda lies in the contextual nature of such content. Symbols and images that signify hatred or extremism in one cultural context may be entirely benign or carry a different meaning in another. "If [the model] never sees an example of a hateful image from a part of the world, then it’s not going to be any good at detecting it," said Jimmy Lin, a computer science professor at the University of Waterloo.
This contest underscores the importance of cultural context in machine-learning models. Existing models often focus disproportionately on US-centric examples, hence the decision by Humane Intelligence to collaborate with a Nordic organisation. "Most of these models are often fine-tuned to US examples, which is why it’s important that we’re working with a Nordic counterterrorism group," said Chowdhury.
However, Lin cautioned that the ultimate solution might involve more than just technology. "We have models that generate fake content. Can we develop other models that can detect fake generated content? Yes, that is certainly one approach," he said. "But I think, overall, in the long run, training, literacy, and education efforts are actually going to be more beneficial and have a longer-lasting impact."
Anticipation is building as the November deadline approaches, with developers racing to devise models that could shape the future of content moderation and digital security. The competition not only aims to provide immediate solutions but also to stimulate ongoing research and innovation in the realm of artificial intelligence and extremism mitigation.
Source: Noah Wire Services