Optimistically - this is now something backpropagation can reward or punish, to make models of any design act smarter. Pessimistically - when a metric becomes a goal it ceases to be a useful metric.