Let’s Design the AI That Kills Us All
Summary
Host Ben Bradford explores the existential risk of artificial intelligence, interviewing AI expert Katja Grace and national security analyst Hamza Chaudhry. They outline a "doom scenario" requiring superintelligent, autonomous agents with misaligned goals and access to destructive weapons. The episode examines why major institutions view AI as a top global threat alongside nuclear war and climate change. Finally, it discusses potential safeguards, including mandatory safety testing, whistleblower protections, and emergency off-switches to prevent catastrophe.
Topics discussed
Intro: AI fears and podcast welcome
Why experts rank AI alongside nuclear war
The three ingredients for a killer AI
Ingredient 1: Superintelligent autonomous agents
Ingredient 2: Rogue motives and goal misalignment
Ingredient 3: Access to doomsday weapons
Counterarguments and expert probability estimates
Current lack of regulation and oversight
Proposed solutions: Testing, whistleblowers, and off-switches
Conclusion: The doom meter and final thoughts
Listen ad-free on Castria