Thank you for confirming what I meant. The crux is learning is learning.
Pirating the book is copyright infringement ... reading in a library is not.
Training of a neural network on copyrighted work is the algorithmic equivalent of a "subjective mental experience".