Human-centered mechanism design with Democratic AI

📖 Description

1 Human-centered mechanism design with Democratic AI Authors: Raphael Koster1?, Jan Balaguer1?, Andrea Tacchetti1, Ari Weinstein1, Tina Zhu1, Oliver Hauser2, Duncan Williams1, Lucy Campbell-Gillingham1, Phoebe Thacker1, Matthew Botvinick1,3 and Christopher Summerfield1,4 Affiliations: 1 Deepmind, London, UK 2 Department of Economics and Institute for Data Science and Artificial Intelligence, University of Exeter, Exeter, UK 3 Gatsby Computational Neuroscience Unit, University College London, London, UK 4 Department of Experimental Psychology, University of Oxford, Oxford, UK *Corresponding author. Email: [email removed] ? These authors contributed equally to this work 2 Abstract: Building artificial intelligence (AI) that aligns with human values is an unsolved problem. Here, we developed a human-in-the-loop research pipeline called Democratic AI, in which reinforcement learning is used to design a social mechanism that humans prefer by majority. A large group of hu...

📊 Game Impacts

💭 Reactions

🔗 Sources

• [1] https://arxiv.org/abs/2201.11441