Faun update: Submitted entry to the Future of Life Institute's worldbuilding contest, consisting of an a history, two short stories, and answers to questions about how the world is going to survive a full-yudkowsky AI alignment problem.
I guess I'll share my entry here about a month from now, when the FLI releases the finalists (unless I don't qualify x-x (unlikely))
Summary tho: Get better proof of danger by querying the AI's thoughts. Use that proof to establish strong global consensus.
Why expect global consensus to lead to strong action, when it hasn't worked for climate change?
Because the proof (An AI straight up saying "I think I'm going to kill literally every human if I'm ever deployed") is a lot less ambiguous, and the stakes are actually a fair bit higher.
But it still might not work.
There are going to be contingency plans in that case, but we shouldn't talk about them.
