An obstruction to capability amplification is a partition of the policy class π into two parts π and π, such that we cannot amplify any policy in π to be at least as good as any policy in π.
[β¦] can we sensibly define βgood behaviorβ for policies in the inaccessible part π?
This seems to be circular, since determining π and π depends on good behaviour being defined.. I guess what is meant is that we amplify policies in π until we hit a ceiling (a fixed point?). Then π = {A β π | Β¬ A reachable from π}. We suspect that π is non-empty, but we donβt know how the policies in there look.
This seems to be circular, since determining π and π depends on good behaviour being defined.. I guess what is meant is that we amplify policies in π until we hit a ceiling (a fixed point?). Then π = {A β π | Β¬ A reachable from π}. We suspect that π is non-empty, but we donβt know how the policies in there look.
(Deleted.)