SUO: Architectures for Intelligent Systems
Answering a message from John Sowa:
http://mars.virtual-earth.de/pipermail/cg/2002q2/000000.html
Philippe Martin wrote :
>
> > A blackboard is an ideal platform for supporting hot-swap or plug-n-play
> > components. When a new component is added to the FMF, it would send
> > a message to the blackboard to identify itself and the patterns of
> > messages it accepts. ...
> > Any server anywhere on the Internet could be converted to an intelligent
> > agent by using an FMF as its front end. ...
>
> What is the main difference between this "blackboard" and knowledge-based
> Web brokers such as the CG-based AI-trader (see the proceedings of ICCS'95 or
> http://www.vsb.informatik.uni-frankfurt.de/projects/aitrader/intro.html),
> the OKBC-based or KQML-based brokers described on the Ontolingua site,
> or the prototype for distributed KBs discussed on the CYC site
> (http://www.cyc.com/applications.html#dai) ?
>
> Knowledge-based Web brokers are ontology/KB servers which, when necessary,
> redirect queries to others ontology/KB/DB servers. Integrating the results
> is an unsolved (and in my opinion unsolvable) problem if the ontologies of
> the various systems are not tightly interconnected.
"tightly interconnected", yes of course, but what does this means?
It does NOT has to mean that each one has to recode ALL the knowledge
from others in order to meaningfully exchange information.
I have been arguing vehemently since Fri 22 Mar 2002 that *incrementality*
should be considered for ontologies interoperability:
http://suo.ieee.org/email/msg08138.html
I got some critiscisms and nit-picking:
mumble, mumble, not gonna be easy:
"can't think offhand how one might implement such a system"
don't ask me to think:
"it is too technical for seweb which is more meant for
general annoucements."
you don't know what you are talking about:
"You might also consult the definition of 'higher-order'
in some standard logic text."
we have no interest beside our (strictly defined) turf:
"It would be better to have a silent list (with hope for
future progress) than an off-topic list"
Etc...
But NO ONE made any comment pro or con *incrementality* per se!
> However, if each (general
> or specialized) KB server mirrors the content of its competitors (manually
> at first, then more and more automatically), the competing servers
> will end up with comparable ontologies/knowledge, and then the processes of
> redirecting queries and integrating results are simpler. Actually,
> if each server manages to include/mirror the knowledge of its competitors,
> the need for a broker is quite reduced: all competing servers will answer
> similarly. That would also be more efficient: large ontologies/KB can be
> handled efficiently but can an application exploiting many distributed KB
> servers be efficient?
Certainly not if this amount to a bulk replication of *all* participating
ontologies!
> I assume that most ontology/KB servers (e.g. WebKB-2, Ontolingua, Ontosausus,
> Ontobroker) could easily include a (very) simple broker, if this is required.
Simplicity of the broker is a well desirable property but scalability
over a very large distributed set of knowledge repositories might
dictate otherwise.
But this is *not* necessarily unavoidable.
What if the broker could be both simple and scalable?
It's all in the design and incrementality might do it!
Anyone to REALLY comment on that?
Rather than on appropriateness in the form of the question or
the subject matter or my credentials, etc...
Cheers.
-- Jean-Luc Delatre
--------------------------------------------------------------
"There is no more common error than to assume that, because
prolonged and accurate mathematical calculations have been
made, the application of the result to some fact of nature
is absolutely certain." - A. N. Whitehead
--------------------------------------------------------------
http://perso.club-internet.fr/jld/ -- GSM: +33 6 11 24 06 29