9 tweets · December 2023 · 55 likes · 4 retweets · read on Twitter
concern: powerful AI systems could spiral out of control into optimizations that would fuck us up. imagine the failure modes of capitalism but - 1000x faster - no human goodwill in the loop literally anywhere - no ability to eg regulate against leaded gasoline or tax cigarettes
= my attempt to distill the center of my AI safety concerns into a broadly accessible tweet
why 1000x faster? it would be able to do things that currently take many humans, with high coordination costs and a bunch of waiting/delegating/communicating/hiring, all in one unit
why no human goodwill in the loop literally anywhere? because the ones that put humans in the loop at all will be slower and outcompeted by the ones that don't
why no ability to eg regulate against leaded gasoline or tax cigarettes? this one is harder to explain but ummm by analogy consider that the war on drugs did not make cocaine disappear
this concern is in magnitude maybe somewhere between 🖇️ the unfathomable turns-everything-into-paperclips scenario 😡 the already-slightly-here fucks-society-up-via-algorithmic-social-media-feed-optimization-fake-news-etc that betterwithout.ai warns about
I am broadly concerned about this. I don't know what to do. It doesn't even require artificial *general* intelligence! hell, maybe having AIs that are general would be better! (far from certain on this)
I don't know how to keep the pressures of capitalism from driving us into such a pit. It may be that personal AIs somehow help individuals come together and defend their values, but I haven't seen a good case for that. share if you have one!
It's possible that there's a way and that we just can't see it from here. It's possible that somebody else sees a way (please share!) tho most of the e/acc folks seem to have written "nothing will go wrong" as their bottom line and rationalized everything else on that basis.