This was mentioned on Slashdot, but it’s worth a read anyway; Clay Shirky on the Semantic Web.
I’ve always been suspicious of the usefulness of Prolog (I’ve seen it used once in my career, as a disassembler-parser generator. Then again, Prolog isn’t used that much in systems programming, so don’t listen to me). Shirky has a very good point; you simply (A) can’t believe the information on the web is complete, or validly-stated, and (B) there is significant hand-waving going on at the wrong levels about making semantics work.
Personally, I think it will always be a “net of a million lies” [Vinge]. Any semantic mechanism is going to have to deal with bad data and outright attacks (a-la search-engine ranking hacks).