Re: CG: Re: Some references about ontology and analogy
On Thu, 06 Jan 2005 15:12:15 -0500, John F. Sowa <sowa@bestweb.net> wrote:
> Danny,
>
> I can't argue against that point:
>
> DA> The Web is the infrastructure on which the Semantic
> > Web is being built, probably the most useful part
> > also being one of the simplest - URIs.
>
> That is essentially what many people were talking about
> in the mid 1980s: extend the Unix hierarchy to the entire
> world. It was a great idea then, and it still is.
The Unix hierarchy didn't get joined up as such (I suppose Gopher was
close) but the servers did get joined up as a graph through HTTP +
URIs. Great idea, as you say.
> I received another offline note with which I agree:
>
> Offline> My impression is that the semantic web is intended
> > mostly, as a massively interconnected named-entity
> > and terminology hierarchy. To improve search engines.
> > Not to do any deep reasoning with. If that is correct,
> > and I think it is, you are arguing against a dictionary.
>
> If that is all that anyone claimed, I would be happy.
>
> What bothers me is that some people kick around words
> like "ontology" and "semantics" without having a clue
> about the subject.
Ok, first let me return to something you said earlier:
[[
I don't believe that more of the same (i.e.,
what Cyc has been doing for the past 20 years or what
the Semantic Webbers plan to do) is sufficient, either
for commercially successful AI systems of for systems
that realistically simulate human intelligence.
]]
I think this is a pretty fair statement. But most people working with
Semantic Web technologies have much lower ambitions. What constitutes
a "commercially successful AI system" is open to argument, but taking
the last point - the majority of SemWeb efforts have nothing to do
with AI in the sense of simulating human intelligence. I think a
reasonable line can be draw between the Cyc aims of giving a machine
human-like capabilities (so it can help humans) and the SemWeb
approach of making human-oriented information more machine friendly
(so it can help humans).
As a side issue, I don't know all that much about Cyc, but it feels
like it aims to implement a thinking machine along the lines of
Asimov-era science fiction. I think one thing we can learn from the
past 10 years or so is that this isn't needed to significantly
augment human capabilities. A system of relatively dumb hosts and
terminals with lots of connections, but with a human at most nodes can
be seen as a society-level augmentation. Regular computer users are
cyborgs, only the HCI is a bit cruder than imagined 40 years ago. But
the whole of connected humanity can be seen as a cyborg too. "We
Robot" if you will.
But going back to the points in the mail you quote - yes, I think it
is reasonable to expect that the majority of the Semantic Web will be
little more than a dictionary. Or rather just a little more than this
- something comparable to what you see in SQL databases. Few people
approach these from the point of view of semantics - they have
queries, they may have a data model (probably largely embedded in
custom objects). The RDF approach provides a more transparent data
model, without translation between the semantics of the objects and
what the store's doing. The model is more declarative. The way the
language is designed allows the free mixing of data from different
domain models and encourages sharing of terminology (and identifiers).
In other words this is relational databases reworked for the Web.
The mention of search engines is telling - I would state the SW answer
as "it's easier not to lose things in the first place". Search is
certainly an application area, but very much search in the more
general sense, for people or places or things, not just documents.
But on the other hand, the layered approach means that it is possible
to use true ontological techniques with this same data as needed
(there are some technical issues between the logics of RDF, RDFS and
OWL, though I believe the majority of the work has been done). There's
a wide spectrum, from the kind of thing we see on the Web now, through
simple data models (which can also be seen on the Web now with
languages like RSS and FOAF) up to fairly complex ontologies (more
comparable to Cyc).
This diversity means that amongst the testimonials for the RDF & OWL
specs [1] you see entries from intelligence/military organisations
(who seem to favour the richer systems) alongside an entry from the
Creative Commons group, which is essentially about simply tagging
content with a machine-readable copyright.
Cheers,
Danny.
[1] http://www.w3.org/2004/01/sws-testimonial
--
http://dannyayers.com