There's a school of thought that the descendants of expert-systems are the key to having auditable decision-making systems, where conclusions come from a composition of rules, and can therefore be explained (and corrected). There are lots of valid criticisms of this approach, but I have yet to see someone offer a compelling alternative.