Superintelligence: Why Smart People Believe What We Cannot Predict

As AI development accelerates, concerns about superintelligence have moved from science fiction into serious discussion among leading technologists and researchers. This talk uses the historical case of nuclear weapons—where scientists could mathematically verify safety margins—to interrogate whether we can meaningfully assess existential risks from machine intelligence. The central insight is that our difficulty in predicting AI outcomes mirrors a broader human tendency to find speculative but intellectually rigorous arguments deeply persuasive, especially when endorsed by prominent figures.