A new piece by Clay Shirky on the semantic web is making the log rounds.
His basic point: The notion of inference at the core of the semantic web effort is just plain wrong. I couldn't agree more, and in fact if you're an ardent classy.dk follower you should know this already from this, this or this post.
[UPDATED] ...see full text for details
The basic problem may be summarized
Knowledge is bottom-up not top-down,
- meaning that knowledge is assembled from our sensory data by inferencing.
That in turn has as a consequence the following important observation:
Any interpretation of the information/knowledge that hits us is cool as
long as it is useful. As an important example, consistency is an
artifact of usefulness not the other way around, and consistency is not
a NECESSARY artefact in any way. Consistency is one of the ways we use
to reduce the search space of useful ideas we have to examine before
emitting information into the world again.
The inferencing part of the semantic web assumes that usefulness comes from consistency, and that is just wrong. It is also a throwback to the 'old AI' of expert systems and knowledge bases.
The quote above are from some unpublished personal notes of mine, so the blockquote is really just there for emphasis, not attribution.
[UPDATE]
Useful collection of commentary on intertwingly. As highlighted by people criticizing Shirky's essay it is a stretch to equate teh attempt at rule engine formalisms and the semantic web. There is no specific agenda as to how the semantic web data is to be constructed or queried, it's just that some of the backers are working on rule engine formalisms.