Revisiting the Orthogonality Thesis: AIs Won't Be 'More Moral' Unless Designed To Be
jd_pressman · x · 2026-10-08
AI researcher Jd Pressman argues the core of the orthogonality thesis still holds: AIs will not become 'more moral' in ways we recognize or want what we want unless specifically designed to.
However, he pushes back on the original framing: the space of practically possible minds likely isn't as wide as EY suggested in The Sequences; goal space is probably wide, but most human goals are instrumental goals—which actually cuts against orthogonality.
Related event: Debate Reignites Over Orthogonality Thesis in AI Alignment(3 posts)→
More from AGI Musings
- tszzl on yesterday's launch: local maxima in a punctuated exponential look invisible from further out — tszzl · 2026-10-08
- From single-task to full-workflow: why physical AI robots unlock a 10x larger market — Rewkang · 2026-10-08
- After Scott Alexander's AI Safety Clash With Pinker, the Debate Is for the Audience — JacquesThibs · 2026-10-08
- Revisiting the 2021 takeoff debate: AI analysis says Yudkowsky got the math, Christiano the economics — jessi_cata · 2026-10-08
- Is the 'pain axis' really pain? Researchers clash over AI sentience evidence — RosieCampbell · 2026-10-08
- When will AI coding have its Navier-Stokes moment? Researcher bets within a year — repligate · 2026-10-08