The correction would be to make the systems learn continuously in an online fashion. This is a hard open algorithmic problem though - how to achieve that in a stable robust way.