Comment by Tarq0n

This is a very basic misunderstanding of machine learning. The goal is always to learn a representation that generalizes beyond the training set without overfitting. If you succeed in generalizing, you can do more than produce derivative work.