Of those four, I think accountability is truly the only moat. The other three are varying amounts of both testable and iterated on via adversarial agents steel-manning the implementations.
Accountability will not come before AIs achieve legal personhood, and that probably will not happen in my lifetime. And if it does in some jurisdictions (which I am not betting money on, this is a full full AGI scenario after multiple more philosophical goalposts move first), I will completely bet it will not be a global recognition.
Accountability will not come before AIs achieve legal personhood, and that probably will not happen in my lifetime. And if it does in some jurisdictions (which I am not betting money on, this is a full full AGI scenario after multiple more philosophical goalposts move first), I will completely bet it will not be a global recognition.