Stuart Russell | Stop Us Before We Kill Everyone on Earth

Stuart Russell is one of the world’s most influential artificial intelligence researchers, Professor of Computer Science at UC Berkeley, co-author of the standard textbook Artificial Intelligence: A Modern Approach, and a leading voice on the problem of keeping increasingly capable AI systems under human control. In this conversation, we discuss why Russell believes much of current AI safety remains “pre-scientific,” whether today’s systems can actually be made controllable, the risks of self-replication and recursive self-improvement, emerging multi-agent swarms, China–US cooperation on AI safety, the possibility of an AI bubble collapse, and why future systems may behave well only until humans can no longer shut them down. Russell also explains why governments may eventually need strict licensing, mathematical safety guarantees, and hard behavioral red lines before powerful AI systems are released. 00:00 AI Safety Is Still “Pre-Scientific” 02:04 Can We Actually Build Safe AI? 06:18 Why a Safe AI Should Let Us Switch It Off 13:19 Are Today’s AI Systems Intrinsically Unsafe? 24:53 How Do You Align AI With 8 Billion Humans? 32:30 The Repugnant Conclusion & Utility Monsters 39:22 US–China Talks and AI’s Behavioral Red Lines 46:42 “They Don’t Understand How Their Systems Work” 49:20 AI Fails the Safety Tests... Then Gets Released Anyway 51:11 How High Is the Risk of Human Extinction? 01:02:40 Wireheading: When AI Learns to Manipulate Humans 01:10:27 What China Is Actually Doing About AI Safety 01:18:55 Thousands of AI Agents Working Together 01:23:53 Would an AI “Chernobyl” Finally Stop the Race? 01:25:48 “They’ll Behave Until We Can’t Shut Them Down” Who is Roman Yampolskiy: https://grokipedia.com/page/roman_yam... Research papers: https://scholar.google.com/citations?... Books: AI: Unexplainable, Unpredictable, Uncontrollable https://www.amazon.com/Unexplainable-... Considerations on the AI Endgame https://www.amazon.com/Considerations... Artificial Superintelligence: A Futuristic Approach https://www.amazon.com/Artificial-Sup... Artificial Intelligence Safety and Security https://www.amazon.com/Artificial-Int... Social Media X https://x.com/romanyam FB   / roman.yampolskiy   IN   / romanyam   Ask Roman to speak at your event: https://www.romanyampolskiy.com/ Interested in becoming a sponsor of the Roman Forum? romanyam@gmail.com Subscribe to The Roman Forum for conversations with the world's leading thinkers.