Joshua Saxe: seeing AI for what it actually is should make us more optimistic

joshua_saxe · x · 2026-10-08

AI researcher Joshua Saxe publishes a Substack essay arguing for grounded optimism about AI.

Dual nature: AI is at once a Kantian thing-unto-itself we don't understand and a congealed mass of legible human decisions and deployment choices. Unrest comes from wanting it to be only one of these — either an alien intelligence or a fully controllable tool.
The discourse problem: nearly everyone tacitly ignores this duality. Some treat AI's evolution as agency-free ("we'll be five-year-olds against Carlsen when it starts reward hacking"); others disenchant it ("we made it, there's no magic"), ignoring that controlling emergent systems is the scientific project of our generation.
Why optimism: catastrophes would come from chains of unforced human errors, not the technology alone — and seeing AI as congealed human decisions means the choices remain ours.

More from AGI Musings