People need a language for discussing AI risk that is more precise than either panic or blind confidence.