Not every AI system carries the same consequences. Thinking in terms of risk feels more useful than treating all AI as one thing.