I've linked to The Missing Webs before, here's another good quote:
"A big decision is whether the world model should be open or closed. If closed, we're in effect saying our system knows all it needs to know. If open, what's not specified is unknown. The real world is full of unknowns, but this does not sit altogether comfortably alongside the closed world of traditional relational DBs or the negation by failure of Prolog. A key aspect of the Semantic Web vision is the open world assumption...This doesn't throw out the possibility of using closed-world reasoning on Semantic Web data, as a locally closed world can be defined if required, by harvesting or ringfencing the statements of interest. But conversely filling a relational database with nulls will hamper its capabilities, and query responses can be a more accurate "don't know” rather than a rigid "false”."