That sounds as if you are imagining a causal graph with arrows from “history” to “evil crazy dictator” and from “history” to “evil”, but not from “evil crazy dictator” to “evil”. So when you cut off the first arrow, this changes none of the causal influences on “evil”. But there should be an arrow from “evil crazy dictator” to “evil”. “History” does not reacharound all the people, to cause “evil” directly, independently of what the people do.
No, but we know already know “evil”. Stuart is suggesting changing one of the arrows that lead up to this node; conditional on our already knowing the value of this node, we need to change other arrows to keep the fit.
You are holding the frequency of evil constant while removing one of its possible causes. Why do that?
The frequency of evil is given to us by the history of the world. Counterfactual considerations give us the conditional probabilities.
That sounds as if you are imagining a causal graph with arrows from “history” to “evil crazy dictator” and from “history” to “evil”, but not from “evil crazy dictator” to “evil”. So when you cut off the first arrow, this changes none of the causal influences on “evil”. But there should be an arrow from “evil crazy dictator” to “evil”. “History” does not reach around all the people, to cause “evil” directly, independently of what the people do.
No, but we know already know “evil”. Stuart is suggesting changing one of the arrows that lead up to this node; conditional on our already knowing the value of this node, we need to change other arrows to keep the fit.