Naive view:
y -> inverse function -> x

Better view:
y -> distribution over possible x