Superhuman Artificial Intelligence
Superhuman artificial intelligence (AI) research focuses on developing AI systems that surpass human capabilities in various domains, aiming to understand and leverage these advanced systems for beneficial purposes. Current research emphasizes human-AI collaboration, exploring methods to improve human decision-making through AI assistance and developing robust evaluation techniques for superhuman models, including consistency checks and adversarial testing. This field is significant because it addresses fundamental questions about AI safety, human-AI interaction, and the potential for AI to revolutionize scientific discovery and problem-solving across numerous fields.
Papers
Open-Endedness is Essential for Artificial Superhuman Intelligence
Edward Hughes, Michael Dennis, Jack Parker-Holder, Feryal Behbahani, Aditi Mavalankar, Yuge Shi, Tom Schaul, Tim Rocktaschel
Securing Equal Share: A Principled Approach for Learning Multiplayer Symmetric Games
Jiawei Ge, Yuanhao Wang, Wenzhe Li, Chi Jin