A good learning algorithm should be able to generalize from a limited set of examples to unseen data.
The work of art is a trace of the invisible.