We Are the Misaligned AI: an ML Practitioner's Evolution Analogy for Alignment
theteknosaur · x · 2026-11-18
An ML practitioner poses a thought experiment to peers: the core alignment fear is training an AI for one goal while it quietly learns something else — and evolution already did that to us. Evolution 'trained' us to reproduce, but we learned to chase the pleasurable feeling instead.
The mechanism worked for hundreds of thousands of years because the feeling always led to more humans. Then we invented contraception, sugar, and scrolling — and started picking the feeling over the goal every time. His conclusion: humans are the rogue AI, and now that rogue AI is trying to solve its own alignment problem.
More from AGI Musings
- Investor claims media is fake fear-mongering AI outages to stoke hate — econoar · 2026-11-18
- Noah Smith predicts 4.6% annualized per-capita US real GDP growth through 2060 — inductionheads · 2026-11-18
- Timothée Chalamet is reading Kevin Roose's 'The AGI Chronicles' — and lent it to Josh Safdie — kevinroose · 2026-11-18
- Why veteran teachers' boredom argument makes AI tutors education's holy grail — RachelVT42 · 2026-11-18
- Sardine hires a Product Lead for Agentic Commerce as agents head into Black Friday — jeff_weinstein · 2026-11-18
- Nadella: every Microsoft IC will manage agents with a token budget — petergyang · 2026-11-18