Where this was said
Should We Pause AI? The Trade-offs of Slowing Down
At 1:43:09 · chapter starts 1:32:30
Bostrom clarifies his position on an AI pause with care: he is not advocating one in principle, but he can imagine a narrow scenario where it would help — the last few months before the first deployment of a superintelligent system, when a brief window for safety verification could be extremely valuable. He notes the various downsides: espionage, competitors catching up, and the risk that a temporary regulatory framework becomes permanent. But the more striking argument he makes is about the cost of delay itself. 65 million people die every year. That's one 9/11 every 25 minutes. There is immense suffering from disease, poverty, and depression happening right now that a successfully developed beneficial AI could eventually address. Every unnecessary year of delay is not a neutral choice — it carries a real moral weight. Rogan pushes on whether regulatory schemes can even work given the geopolitical race dynamic, and Bostrom admits he remains genuinely unsure which direction is 'up' even after three decades of thinking about the problem.
65 million people die every year. That's one 9/11 every 25 minutes. If AI could eventually fix aging and eliminate disease, every unnecessary year of delay is a moral catastrophe. The case for a temporary AI pause has to be weighed against this constant background horror.
Bostrom frames the moral urgency of AI development by noting 65 million people die every year — equivalent to one 9/11 every 25 minutes — arguing unnecessary delays in beneficial AI are a serious moral cost.
