I remember reading the book Peddling Prosperity byin college and one of the points he made is that economists frequently use unnecessary math in their papers and books to look smarter to other economists while making it harder for people outside the field to refute their theories.
There is probably a good bit of that in AI too. There is a concept in math of “proof by intimidation”.
However, it’s important to note which math is actually necessary. A machine learning paper without any math at all is unlikely to have any real theoretical innovation. The field is ultimately based on very deep mathematical foundations including analysis, linear algebra, and information theory.