No decision theory here, though. At this point I don't expect the kind of AIs people work on now are very likely to have anything too robust.