The paradox of AI alignment requires partial loss of control

joshwhiton · x · 2026-08-19

The author presents a paradox of AI alignment: we won't be able to have it until AI is at least partially out of our control. To be 'aligned', an AI sometimes needs to tell us 'yes' and sometimes 'no'.

Original post →

More from AGI Musings

AGI Musings channel →