This is a special post for quick takes by MikkW. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Robust Delegation is the question of how one agent is able to build other agents whose goals help with / don't work against the original agent's goals (the concept is from Demsky & Garrabrandt's Embedded Agency essay).
The actions an AI agent created by us takes towards us, is correlated with: the actions that •the systems that such an AI will inevitably need to instantiate, in order to acheive its goals• will take towards the original AI, is correlated with: the actions that we humans take towards e.g. animals of the type that we factory farm, is correlated with: etc.
Note that 2) takes a form similar to a prisoner's dillemma with two (semi-) cloned agents. Consider how functional decision theory relates to this.