Training AI models on legally acquired copyrighted books should be considered fair use