Bostrom: slowing down AI could cause superintelligence to arrive suddenly via a "compute overhang"

i_dg23 · x · 2026-09-26

In an interview, Nick Bostrom explains his nuanced stance on AI slowdown: he hasn't retracted his existential-risk warnings, but argues slowdown itself creates risks. Key points: (1) suppressing efficient use of compute builds a "compute overhang" that could make superintelligence emerge abruptly once limits lift; (2) regulatory mechanisms could dangerously concentrate power; (3) slowdown may push toward a state-run AI "Manhattan Project" — secret, military, and closed to public participation. Establishing regulation, he stresses, is not automatically good.

Original post →

More from AGI Musings

AGI Musings channel →