Modern Wisdom

Podbit · Modern Wisdom

The Collapse of American Politics - Ezra Klein - #1114

Explore episode Jun 22, 2026
Technology
AI Safety Is Stuck in the Future While the Present Needs Governing

The Collapse of American Politics - Ezra Klein - #1114 · Jun 22, 2026 Technology

If recursive superintelligence slips out of our control overnight, we're probably just fucked — no regulation will catch that. Klein's argument is that the AI safety movement has been paralysed by speculative worst-case scenarios while neglecting the powerful AI systems that already exist and already need democratic governance.

Where this was said

The Urgent Problem of AI Safety

At 1:10:00 · chapter starts 1:09:00

Williamson, who first read Nick Bostrom's Superintelligence in 2014-15, sets up the existential risk framing and asks Klein for a 'white pill.' Klein offers something more nuanced: genuine concern about AI risk, a non-trivial P-Doom, and a critique of how the AI safety debate has been structured. His key argument is epistemic: you cannot solve a problem whose shape you do not know, and we have spent too long in a speculative mode that cannot be acted upon. The 'fast takeoff' scenario — recursive self-improving superintelligence that seizes power overnight — is probably wrong about how intelligence translates into power. As Dwarkesh Patel has argued, the capability to wield power is much more than intelligence; the world is full of friction. Even Stalin was not the world's greatest genius. Klein praises Chris Olah's mechanistic interpretability work at Anthropic as exactly the kind of research that should be funded — building genuine understanding of these systems — but argues that the political system needs to stop debating the absolute worst thought experiment and start building the regulatory competency to govern AI in the present moment.

Similar podbits