Re: [tlug] Effect of AI and data processing rift on the Semantic Web

["Stephen J. Turnbull" <stephen@example.com>]

Edward Middleton writes:

 > Using your perception of the situation to avoid investigating it.  If 
 > this is not the case could you explain specifically what you see as 
 > being OSI-ized.

RDF, for example, is a steaming pile of compost.  Oh, it will work
(and obviously does) as a machine-readable language, but like many
applications of XML it's butt-ugly and internally inconsistent (eg,
the text-based "^^" convention for adding type information to values).
The stuff built on top of it (OWL, FOAF) is just as bad AFAICS.

I mean, the immediate response of any sane person reading the RDF
Primer provided by the W3C is going to be "God, no!  Not unless my job
depends on it, and even then I'll take a 20% pay cut to avoid it."

This matters, because it's *people*, not *machines*, that for the
foreseeable future are going to be making connections between FOAF and
the Dublin Core, for one example.  Making them work in RDF is just not
a good way to get them excited.

 > The views you have expressed seems to be regularly espoused on the 
 > semantic-web mailing list but it seems to be almost universally by 
 > people without even a layman's understanding of knowledge 
 > representation[1] and computer reasoning[2].

Argumentum ad hominem.  When specialists are accused of wanking, they
always resort to the claim that laymen don't know the difference
between a bird in the hand and two in the bush, at least when it's a
specialist's bird.

Of course those with the specialities you mention will consider the
semantic web activity meaningful.  It's not an interesting piece of
evidence.

But I have to wonder if these specialists really know what they are
doing, even.  Take the RDF "^^" convention.  RDF is clearly a model of
category theory, since it concerns graphs and composability of arcs
between typed sets.  But the "^^" makes it impossible to use category
theory to analyze RDF on its own terms.  Instead, you need to
translate RDF as standardized by the W3C into RDF-- which instead of
using "^^" uses more precision in the labels (ie, the value itself
knows what its type is).

 > > For heaven's sake, Berners-Lee invented the semantic web in 1994!
 > > In some areas, "web ontologies" are taking hold, such as the
 > > Dublin Core and RSS feeds. But until they start to connect up in
 > > useful ways, there is no "the" semantic web.  That's what makes
 > > the World Wide Web such a wonderfully concrete thing: the
 > > fundamental facts that the Internet is a singleton and that any
 > > URL on the Internet can be accessed from anywhere on the
 > > Internet.  It's all connected, by construction.
 > 
 > Compared to developing the WWW, developing a Semantic Web is a 
 > non-trivial problem.  Not because it is difficult to put semantics in 
 > web pages, but because it is difficult to work out what semantics need 
 > to be added to allow AI agents to reason, plan, learn and perform other 
 > useful tasks.

What I'm saying is that trying to work them out from a priori
conceptions is OSI-izing.

Eg, we'd like to connect literary objects (Dublin Core) to authors
(FOAF?) and updates (RSS).  I think that you would agree this is an
interesting component of a universal semantic web (if not, try
proposing one yourself and we can discuss that one instead).  Who's
working on that problem?  What technology are they using?  How would I
find the answers to those questions?  Of course, answering these more
general questions is something we'd hope a universal semantic web
could help to do.

 > As you say the WWW has its roots in Project Xanadu started in
 > 1960[3] the semantic web in 1994[4], considering this, its progress
 > has been pretty impressive.

*I*'m not impressed.  Show me something impressive.

P.S. Your MUA allows footnotes to get out of sync.  No semantic web
technology there, I guess.  ;-)

 > They are basically providing a means to  use "what works" in AI to web 
 > problems.   Without some sort of interfaces and implementations using 
 > "what works" in AI isn't a *practical solution*.

Practical solution to what?  The rosy palm problem of semantic web
specialists?

 > There are an infinite number of problems on the net that lend themselves 
 > to existing AI approaches, it doesn't make sense to treat them each 
 > individually with a proprietary solution when you can attack a whole 
 > class of similar problems.

Well, attack, yes, but what did they attack?  IMHO, RDF for one
example is a great gaping smoking hole in the W3C's collective foot.

 > > I don't which class FOAF falls into.  Maybe FOAF is as common in some
 > > reasonably large circle as meishi-koukan is in Japan.  If so, it
 > > solves a practical problem, and I don't need to know what it is,
 > > either.  Otherwise ....

 > FOAF[6] is an ontology implemented using W3 OWL.  It solves a class of 
 > problems relating to identity management.

I know how to use Google.  For example, without looking again I can
tell you that ISTR the FOAF page claims FOAF is a "vocabulary", not a
full "ontology".  Where in practice is this solution being used?  For
example, does OpenID use it?  Can you give me a similarly important
example where FOAF is a solution?  Or one where lots of people use it?

 > Look at Googles Social Graph[7] for some of the potential uses.

I don't want "potential".  I want to see proven value.  The WWW was
demonstrating proven value within a couple of years of its inception,
despite the very poor penetration of the Internet itself.  The
semantic web has had 50 years of dreams since Ted Nelson and 15 years
of committee meetings since Tim Berners-Lee realized that semantic
links were missing from the WWW.  There should be some corners of the
Internet where semantic web techniques are producing such value.