Ramez Naam: no strong evidence of instrumental convergence in superintelligent AI

ramez · x · 2026-09-05

In an X exchange on AI safety, Ramez Naam argues that agentic superintelligent AI is inevitable for economic reasons, but there's little sign of strong instrumental convergence — goal drift, power seeking, or self-preservation. His point: those traits were selected by Darwinian evolution, and the design space is wide enough that we may be able to design them away. Interlocutors push back that control itself becomes moot once AI is smarter than you. The thread follows recent safety concerns inside and outside the labs.

Related event: Debate Erupts Over Instrumental Convergence and Superintelligence Alignment(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →