The value of information is not in knowing more, but in knowing what others don't know.
A good learning algorithm should be able to generalize from a limited set of examples to unseen data.