Someone might argue that the simple algorithm for a paperclip maximizer in the previous post ought to work, because this is very much the way currently existing AIs do in fact work.