Decision scoring without output rationales poses unacceptable risks for automated AI systems