Principles of Content Addressing

This is an attractive idea. But wouldn't the large space of all documents on the web likely lead to hash collisions?