Mistakes were made? A critical look at how EA approached AI safety | David Krueger | EAG London 23

David Krueger discusses his research on deep learning, AI alignment, and AI existential safety. He examines the AI safety community's approach to the issue, highlighting the limitations of a techno-solutionist mindset. Krueger emphasizes the importance of collaboration with academia and civil society, and the trade-offs of working with big tech companies. He explores the concept of AI safety in terms of structural risk and expresses concern about individuals' negligence in understanding existential risks. The chapter also delves into the challenges of predicting threat models and discusses the unfolding of AI's existential risks. Krueger suggests the idea of regulating compute in AI systems and emphasizes the need for a global consortium to control advanced compute's supply chain. He also explores the concept of emergent agencies in AI algorithms and the importance of AI governance research and informing policies.