> If AI destroy the world who would make them accountable?
You got that backwards. The point is that not rewarding them for their putative efforts to destroy the world makes that outcome less likely.
> If not, then the risk is fictional and doesn't matter.
Again, the same mistake - destroying the world and Ant's actions aren't independent random variables, the latter does matter for the manifestation of the former.
You got that backwards. The point is that not rewarding them for their putative efforts to destroy the world makes that outcome less likely.
> If not, then the risk is fictional and doesn't matter.
Again, the same mistake - destroying the world and Ant's actions aren't independent random variables, the latter does matter for the manifestation of the former.