It’s true that agent() cannot see where world() ends and agent() begins, but we can see that! Otherwise it would be unclear what a “decision-theoretic problem” even means. So we can just compare the return values of world() for different implementations of agent(). Or have I mis-parsed your question?
It’s true that agent() cannot see where world() ends and agent() begins, but we can see that! Otherwise it would be unclear what a “decision-theoretic problem” even means. So we can just compare the return values of world() for different implementations of agent(). Or have I mis-parsed your question?