Yeah yeah, in this we agree. I was just lazy. One should say: X is a random variable representing the choice of a message at random (under probability distribution PX). Then h(X) is a random variable, that induce a different probably distribution bla bla.