You might have come across Judea Pearl's new book, and a related interview which was widely shared in my social bubble. In the interview, Pearl dismisses most of what we do in ML as curve fitting. While I believe that's an overstatement (conveniently ignores RL for example), it's a nice...