I am of opposite opinion. I think it is unreasonable to train AI using copyrighted information without permission from the copyright holder, at least if it is done to create a proprietary product. It is probable even unreasonable if the end result would not be proprietary but at least that would benefit the world more.