My conjecture is that you cannot fully separate MA and AI safety / alignment—or worse, solve AI safety first and then ask AI to solve values for you. We should solve them together, as some sets of values will be incompatible w some approaches to safety, and some AI development pathways will make some sets of values inaccessible (e.g., I don’t think that an egalitarian world for our descendants is a likely outcome w the current trend)
My conjecture is that you cannot fully separate MA and AI safety / alignment—or worse, solve AI safety first and then ask AI to solve values for you. We should solve them together, as some sets of values will be incompatible w some approaches to safety, and some AI development pathways will make some sets of values inaccessible (e.g., I don’t think that an egalitarian world for our descendants is a likely outcome w the current trend)
Yes I completely agree, Moral alignment and controllability/safety alignment are very interconnected and one effects the other