A proper solution to this problem would be an optimal decision theory. Consider the decision itself as a random variable, then take some epistemic model of the world and infer, from a third-person point of view, what an optimal agent with certain knowledge and preferences should have done. Then output that decision.
A proper solution to this problem would be an optimal decision theory. Consider the decision itself as a random variable, then take some epistemic model of the world and infer, from a third-person point of view, what an optimal agent with certain knowledge and preferences should have done. Then output that decision.
I am not talking about optimal at all. Just being able to forensically detect any sign of agency rather than… what?