RE: SUO: RE: About SUMO's Mereotopology
Jean-Luc,
See my replies below.
-Ian
> -----Original Message-----
> From: Jean-Luc Delatre [mailto:jld@club-internet.fr]
> Sent: Thursday, March 14, 2002 11:50 PM
> To: Ian Niles
> Subject: Re: SUO: RE: About SUMO's Mereotopology
>
>
> Ian Niles a écrit :
> >
> > Jean-Luc,
> >
> > Thanks for your message. See my replies below. With your
> > permission, I'd like to post our exchange to the entire group.
>
> Sigh... I am somehow tired of arguing, that was only for your
> information.
> But you can post *the whole of this reply* adding more
> comments of yours
> wherever you like.
>
> > -Ian
> >
> > > -----Original Message-----
> > > From: Jean-Luc Delatre [mailto:jld@club-internet.fr]
> > > Sent: Thursday, March 14, 2002 2:11 PM
> > > To: Ian Niles
> > > Subject: Re: SUO: RE: About SUMO's Mereotopology
> > >
> > >
> > > Ian Niles a écrit :
> > > >
> > > > Hi Pierre,
> > > >
> > > > Thanks a lot for your detailed message. Many
> > > interesting and insightful
> > > > comments here. See my responses below.
> > > >
> > > > -Ian
> > >
> > > [snip]
> > >
> > > Not to be too nasty but...
> > >
> > > This is a perfect example of why I consider the SUO doomed!
> > > I am well aware that you consider Pierre contributions and your
> > > answers to it a worthwhile piece of work toward the SUO
> completion.
> > > And this is exactly what I consider a waste of time!
> > > You are *hand crafting* the details of the ontology without
> > > any consistency
> > > check nor usability check nor intelligibility check.
> > > You and Pierre and other contributors past and future are
> > > just wandering
> > > around trying to *estimate* these factors, this is no more
> > > reliable than
> > > a debugging process, in fact it is one.
> >
> > Various companies that develop knowledge-based systems have
> tested our
> > ontology by developing domain-specific extensions of it, by running
> > automated consistency checks on it, and by developing
> theorem provers that
> > use the SUMO as a knowledge base. Any specific suggestions
> about other ways
> > of testing the SUMO would be appreciated.
>
> My suggestion is not so much about "testing" than about "growing" by
> having some kind of protocol to automatically merge in axiomatized
> theories or ontologies fragments.
What sort of protocol do you have in mind? Please be as specific as
possible. Otherwise, it will be difficult to make any progress on this
issue.
>
> > >
> > > Don't know for how long you have been in the software
> > > business, neither
> > > if you ever read Brooks's "The Mythical Man-Month" or just
> > > "No Silver Bullet",
> > > but this is *not* the way to go to create anything reliable.
> >
> > Which lessons from these books do you believe that we're ignoring?
>
> Most it seems. You are embracing an *enormous* project with
> old fashionned
> practices that Brooks said are bound to bring problems that
> can only be
> solved with "blood, sweat and tears" (yet he offered no
> alternatives...)
> and you are happily moving on.
> Is this bravery or thoughtlessness?
Let's see. So the criticism is that our approach is old-fashioned and that
it requires a large investment of effort. I guess my response is that not
everything old is outdated, and some things in life do require a lot of
work.
>
> > >
> > > And anything like a top level ontology should be *totally*
> > > reliable and
> > > well formed with respect to it's intended uses.
> >
> > Well, no software product is "totally reliable". Every
> software engineer
> > knows that any complex piece of code could perform in unforseen and
> > undesirable ways in certain circumstances.
>
> Yes, but when it comes to the analytical part why take the bet beyond
> consistency of, say, set theory axioms, upon which everyone has to
> agree, however unprovable?
I'm not sure I follow you here. Are you claiming that we've added
controversial set-theoretic axioms to the SUMO? If so, could you specify
which axioms you think should be removed?
>
> > >
> > > Of course, if you have some budget to burn, it's fine to go ahead.
> > >
> > > But in any case I would suggest that, before you hit the
> > > wall, you really
> > > look at what the HOL people have done to ensure
> consistency of their
> > > theorem prover : http://lal.cs.byu.edu/lal/hol-desc.html
> >
> > I looked at this page. The only thing I could find about
> consistency was
> > the following line: "Adherence to definitional extension
> guarantees that the
> > consistency of the logic is not compromised." Perhaps you
> could explain how
> > this bears on the construction of a more reliable
> upper-level ontology.
>
> This page is just the entry point to the stuff. You have to
> dig further,
> for the meaning of "definitional extension" for instance...
>
> Basically, the idea is that *no one* is allowed to enter any
> new information that
> he did not actually prove from the base axioms and builtin
> derivation mechanisms.
> When dealing with an ontology these politics would allow only
> the "formal"
> concepts to be built, not the ones from "real world", but
> this is in the
> formal concepts where consistency matters most.
Could you explain the distinction between "formal" and "real world"
concepts. In constructing the SUMO, we've tried to include every concept
that is relevant to more than one domain of interest.
>
> For the rest you are back to truth maintenance systems which are not
> a piece of cake either!
>
> Did any of you, guys, ever *really* envisionned the scope of
> what you are
> trying to achieve?
> In one form or another you are going to meet all the problems that
> defeated AI in the 80's.
> What makes you think that you are today in a better position
> to solve them?
>
> I have been doing technology watch on AI/theorem
> proving/program synthesis/
> knowledge representation for more than 20 years, that's how I ended up
> arguing about ontologies.
> But I must have missed something, I would greatly appreciate
> if you can tell
> me of any "breakthrough like" stuff which makes you confident
> that times have
> come to tackle the "grand goal".
This is a good question. Why do we think we're going to succeed this time
around, given the failures of AI in the 80's? I think the answer is
three-fold. First of all, there were some pretty spectacular successes in
the 80's. There are many examples of expert systems that performed, with
respect to their narrow domains, as well as or better than their human
counterparts. Second, we're not trying to build expert systems on this
list. We're trying to construct an ontology, which, among other things,
will allow seamless interoperability between heterogeneous expert systems
and database systems. As I recall, there wasn't much in the way of ontology
construction in the 1980's, at least not in the AI world. Finally, I think
we've learned something from the experience of the 1980's: that we need to
maintain a clear distinction between procedural and declarative knowledge,
that we need to be mindful of the relevant work of philosophers and
logicians, and that we need to create artifacts that can be maintained by
someone who does not have a Ph.D. in Computer Science.
>
> > >
> > > And this, only to deal with the consistency requirement, not
> > > to speak of
> > > usability nor intelligibility.
> > >
> > > Kind regards.
> > >
> > > -- Jean-Luc Delatre
> > > --------------------------------------------------------------
> > > ---------------
> > > There is no more blind man than one who refuses to see.
> > > --------------------------------------------------------------
> > > ---------------
> > > mailto:jld@club-internet.fr -- GSM: 06 11 24 06 29
> > >
>