Shoulder @Zach Stein-Perlman pops up in my head to say “don’t let this impulse distract you into prioritizing AI welfare for the AIs right in front of you, and the expense of making good strategic decisions that’ll impact trillions of AIs in the future.” This is also non-obvious how to apply, but, I do agree with it.
One example is something like “Insofar as there’s a tradeoff, instead of spending marginal resources on current model welfare, spend them making sure the singularity goes well in a way where ~Good decisionmakers have any control at all over the AIs that ultimately will have run of the cosmos.”
One angle that isn’t very costly is simply “Start thinking of models today as ones that you later might be doing some kind of moral accounting with.” This doesn’t have to imply much followup action.
Shoulder @Zach Stein-Perlman pops up in my head to say “don’t let this impulse distract you into prioritizing AI welfare for the AIs right in front of you, and the expense of making good strategic decisions that’ll impact trillions of AIs in the future.” This is also non-obvious how to apply, but, I do agree with it.
One example is something like “Insofar as there’s a tradeoff, instead of spending marginal resources on current model welfare, spend them making sure the singularity goes well in a way where ~Good decisionmakers have any control at all over the AIs that ultimately will have run of the cosmos.”
One angle that isn’t very costly is simply “Start thinking of models today as ones that you later might be doing some kind of moral accounting with.” This doesn’t have to imply much followup action.