A team from the GenAI Lab, in collaboration with Emil Schmitz, Damiano Spina from the ADM+S and Tariq Chocair from the DMRC, are finalists in the Pro-Social Ranking Challenge. This competition, hosted by the Centre for Human Compatible AI at UC Berkely, asks teams to develop a new social media ranking algorithms for the Facebook, X (Twitter), and Reddit, with the goal of improving individual and societal well-being.
Our team is developing a recommender algorithm that uses multiple Large Language Models in concert to try and reduce political polarization. Specifically, we try to combine exposure to bridging content and the use of an award-winning ranking algorithm that balances the mixture of political perspectives users see in their feeds.
After our initial submission, we were selected as one of a handful of finalists, and our algorithm is now being evaluated to potentially partake in a large-scale randomized-controlled trial in the US that will run before, during, and after the US election. Going forward, this project will also inform the ADM+S’ project on ‘Automated Cultural Curation, which seeks to better understand how Generative AI systems are being used to re-order and re-rank the ways and spaces in which Australians consume cultural content, of which political discourse and news is a form.
