Assurance of AI Systems From a Dependability Perspective