It’s true that restricting what AI can train on inhibits societal progress, but it’s consistent with current copyright laws because an AI is not a human and can’t be treated as anything more than an algorithm. What we’re learning here is that AI is bringing to light a problem in intellectual property law that’s plagued us for a long time: intellectual property being overly protected is harmful to society as a whole. I wouldn’t be opposed to AI training on data on the internet if people got the same treatment: let people reuse each other’s melodies, don’t protect likenesses so strictly, and for the love of humanity, no more pharmaceutical patents!