A good learning algorithm should be able to generalize from a limited set of examples to unseen data.
The act of writing is an act of optimism. You would not take the trouble to do it if you felt it didn't matter.