> Buying used copies of books, scanning them, and training on it is fine.

Awesome, so I just need enough perceptrons to overfit every possible copyrighted works then?