Re: Semantic Web issues

[Charlie Abela]
> I was recently involved in a discussion about the semantic web with a
friend
> of mine who is rather sceptic about some ideas that are being presented on
> the topic. Some of his arguments were quite convincing.
>
> His view of the Semantic Web (of which Web Services is a specific
sub-part)

Well, I do not consider the currently hyped notions of "Web Services" to
have much to do with the Semantic Web.

It is possible to use many or maybe all of the technologies being talked
about for the Semantic Web in completely non-SemWeb ways as well.

> is that it is primarily (if not practically exclusively) focused on
> expressing the meaning of information as the author intended. Except that
he
> believes that it is extremely limiting. He argues that discoveries are
made

This would depend on what you mean by "discoveries".  If I want to find a
doctor in my area that treats allergies, it is quite a different thing from
discovering relationships between, say, cataloging of publications in a
library and hash tables.  Again, both of these are different from
"discovering" evolution.

> by taking data and interpreting it in an unexpected 'context'  (whatever
> context means) and that the approach that is being considered by the
> Semantic Web community to information representation and reasoning may
> prevent information discovery.
>

Could you say more about what you (or your friend) think the SemWeb
"Community" approach is?

> Reasoning in the Semantic Web is monotonic and makes an open world
> assumption, rather than nonmonotonic and making a closed world assumption.
> He is reluctant to believe that taking an open world, monotonic approach
to
> reasoning will necessarily ensure that the information
transmitter/receiver
> will actually be able to work out that they are "talking" about the same
> thing without first decontextualising the data being reasoned with/about,
> which he argues that in itself this is intractable.
>

Not too sure what you mean by "decontextualizing".  Usually, you have to
have context for anything complicated because you cannot specify enough
specifics.  The rest has to covered somehow by context.  So
"decontextualizing" sounds questionable.

Yes, the problem of identity is a tough one, one of the hardest, it seems to
me.  But just because you have the SemWeb does not mean that all systems
have to be able to work out the identity of all things known to all other
systems.  Just like you can have an ontology pieced together from a few bits
for some particular purpose (rather than having to have a global ontology
before you can do anything), you can get useful things done without having
to be global in scope.  At least, that is the hope.  FOAF is a sort of tiny
test case in this area, and seems to be working out well so far, small
though its scope is.

> He is also concerned about the approach taken by the Semantic Web that
> assumes that information (data in context) is consistent,

I do not think that the plan is to assume everything is consistent.
Naturally, that is the easiest way to proceed, so a lot of people are going
to want to get started without worrying about inconsistencies.  But if you
read the newer versions of the RDF and OWL specs, for example, you will see
that there is a lot of talk that ultimately relates to dealing with
inconsistencies.  For example, instead of having a fault condition because
of a contradiction, or proving something incorrect, you may find that there
is no interpretation that satisfies all the recorded statements.  That is a
different thing altogether.

Will this kind of thing be practical?  It remains it be seen, but it seems
to be very much in the thoughts of most of the top players.

> because
> i) this is only possible with undisputed/indisputable facts (which are
only
> a proportion of the 'information' humans use to reason with/about), and
> ii) will it necessitate a Microsoft-like company to make available (for a
> fee, of course!) consistent information for open world, monotonic
reasoners
> to use?
>

This assertion is more of a closed-world old-style AI idea, and it does not
match up all that well with current-day discussions of RDF, OWL, tableaux
reasoners, etc.  OTOH, good ontologies are hard to create, and it would not
be surprising if some of them were to be controlled by large commercial
interests.

There is plenty of room for questioning the establishment of identity (as
you point out).  Another questionable area is whether a triples model (i.e.,
RDF) in which statements do not have identities of their own can be powerful
enough to handle what will be required.  Another is the practical behavior
of reasoning systems with large amounts of incomplete or contradictory
information., including processing time.

But these do not seem to me to be the kind of things you were questioning
(well, identity is, but the real question seems to be whether a thing can be
identified well enough by its properties and relationships that the
identification becomes reliable.  This certainly remains to be seen as a
practical matter, but people do it all the time, so maybe software can too.

Cheers,

Tom P

Received on Friday, 18 July 2003 02:12:02 UTC