Scott Alexander's sprawling, essential essay on coordination failures — why systems that nobody wants emerge anyway, and what that implies for everything from capitalism to AI alignment.
“The essay I find myself referencing more than any other. Moloch as a concept for 'the thing that makes systems defect even when everyone inside them would prefer they didn't' is extraordinarily useful.”
2 comments
Join OpenLinq to join the discussion
The section on the 'race to the bottom' being structurally incentivized — not a choice — is what I keep coming back to. It's not that people are bad. It's that the game is set up wrong.
Reading this alongside the scaling hypothesis essay changes what you think AI alignment is actually about. The problem isn't making AI 'good' — it's making sure Moloch doesn't get into the weights.