I posted this question over at math stackexchange after the thread about category theory and haskell from yesterday[1].
I found Carl Mummert's answer especially illuminating and definitely recommend reading it. He compares category theory to a strongly typed language, and set theory to a weakly typed one.