Well, wouldn't the same be true for using language?  Language follows a grammar, uses established vocabulary based on an underlying set of characters or speech forms; why would anyone follow the rules?

Formalists in mathematics argue that mathematical structures are very much conventions, which could perhaps be otherwise*, but there are certain quite widely used conventions that have shown value in our studies of the world and/or of mathematical practice itself.  Learning the Peano or Zermelo/Frankel axioms is acquiring a very useful vocabulary for the practice of a certain way of number or set theory, and we say this encapsulates a productive way of thinking about things without demanding that they are the only right way to learn to multiply or to perform abstract operations on collections of objects.

\* *indeed, alternative axiomatizations or even alternative axiom systems are often subjects of independent mathematical interest*