I think competence-sorted social strata are incredibly important for aligning higher-competence people with actually doing what’s good and seeking what’s actually true. If you use the simple model that people will behave according to what they predict other people will judge as good behaviour,[1] and what they will judge as good behaviour (at least in EA, ideally) is that which they predict will help others the most… then for people who are really good at figuring out what behaviour actually help others the most, they will only be motivated to do those things insofar as they’re surrounded by judgers who also are able to see that those behaviours are good.
So if you have no competence-sorted social strata, the most competent people will be surrounded by people of much lower competence, which in turn means that the high-competence people won’t be motivated to use their competence in order to figure out ways of being good that they know their judgers can’t see. On this oversimplified model, you only really start to harvest the social benefit from extreme competence once you have all the extremely competent people frequently mingling with each other.
This is why I personally am in favour of EAs trying more (on the current margin) to hang out with people similar to them, and worry less (on the current margin) about groupthink. We’re already sufficiently paranoid of groupthink, and a few bubbles getting stuck on crazy is worth the few bubbles bootstrapping themselves to greater heights.
I think competence-sorted social strata are incredibly important for aligning higher-competence people with actually doing what’s good and seeking what’s actually true. If you use the simple model that people will behave according to what they predict other people will judge as good behaviour,[1] and what they will judge as good behaviour (at least in EA, ideally) is that which they predict will help others the most… then for people who are really good at figuring out what behaviour actually help others the most, they will only be motivated to do those things insofar as they’re surrounded by judgers who also are able to see that those behaviours are good.
So if you have no competence-sorted social strata, the most competent people will be surrounded by people of much lower competence, which in turn means that the high-competence people won’t be motivated to use their competence in order to figure out ways of being good that they know their judgers can’t see. On this oversimplified model, you only really start to harvest the social benefit from extreme competence once you have all the extremely competent people frequently mingling with each other.
This is why I personally am in favour of EAs trying more (on the current margin) to hang out with people similar to them, and worry less (on the current margin) about groupthink. We’re already sufficiently paranoid of groupthink, and a few bubbles getting stuck on crazy is worth the few bubbles bootstrapping themselves to greater heights.
I think it goes one meta-level up from this, but let’s not needlessly complicate things, and this level is predictive enough.