It occurred to me recently why the "semantic web" (basically a computer-readable meaning-encoding scheme so that people could indicate to the computer what the english that they wrote was about and the computer could then facilitate intelligent searching) is inherently doomed to failure: to be really useful, one would have to encode as much information in the semantic web markup as one did in the english — i.e. after writing the document in English, one would then have to go and translate it into semantic web markup. Translating is often nearly as difficult as composition.
Thus any attempt at the semantic web (where the computer-readable semantics are created by humans) will fail either because it's hard enough for most people to write something once — writing literally everything twice is just way too much work. I believe that this is an inherent flaw in semantic management — the correct solution, in any form, has too high a minimum amount of work.
I suspect, too, that it's a pretty steep drop-off from there; if the computer doesn't actually understand what's being said, the usefulness probably plumets. This means that it's likely that text searches like what google provides will probably be the best which is achievable without artifical intelligence (in the sense of something which can understand abstract concepts and abstract relationships).
(Of course, if we ever invent an artifical intelligence, it's hard to imagine why it won't get as bored with being a search engine as human beings would be, but that's a different topic entirely.)