RE: SUO: Re: Proposed SUO Content Outline
>John Sowa wrote:
>
> > I would like to do that. But just one point about KIF: I don't
> > consider KIF an aid to communication between humans -- on the
> > contrary, the "merged ontology" would be vastly easier for all
> > of us to understand if it were documented in the way I have
> > tried to document the underlying concepts in my paper on
> > processes and causality.
> >
> > That doesn't mean I am against using KIF. But I want to strike
> > a balance between two different goals:
> >
> > 1. Formulating our ideas in a form that is comparable to the
> > usual mathematical textbooks, which interleave each formula
> > with a great deal of discussion directed toward human
> > readers rather than computers.
> >
> > 2. Translating the resulting formulas into a computable
> > notation, such as KIF, CGs, DAML, OIL, RDF, or whatever.
> >
> > Of these two activities, the first is much harder, much more
> > important, and much more time consuming. Once the first task
> > is completed, the second is trivial in comparison. (But I also
> > recognize the importance of doing both tasks in parallel and
> > iteratively -- start with #1, do some of #2, test the results
> > on a computer, and repeat with a refinement and extension of
> > #1 and then #2 and back again.)
>
>Can I support the wisdom of what John has written here as it seems to me to
>be crucially important. I cannot begin to count the number of times when
>looking at data models and other formalisms that a lack of appropriate words
>has rendered the computable notation free of meaning (in the sense that no
>useful connection between the model and the world is possible) or has made
>its rationality deeply suspect.
But let me sound a different note. While I agree with John and Chris
that adequate, and even profuse, documentation is important to convey
to human users what the axioms of an onotology are *intended* to
mean, the fact remains that the only actual content in a formal
ontology lies in the axioms. The ONLY content. It is perilously easy
to assume that because one has written a convincing essay which
explains what the concept of, say 'physical object' is that one has
in mind, that this meaning - or at any rate, some smidgeon of this
meaning - is thereby somehow magically captured in one's axioms just
by writing (PhysicalObject ?x). This peril is well known in AI
circles, where it has become known as the 'gensym fallacy', and we
take pains to burn it out of our students heads at an early stage of
their education. One way is to imagine the entire axiom system with
every meaningful name replaced by something like "G2456734" (a
'GENerated SYMbol'), and then see what the axioms mean. To a
mechanical reasoner, they mean *exactly the same as they did before*,
no more and no less. To think otherwise is like thinking that a
compiler is reading and understanding the comments in your code.
A related fallacy is the idea that by using 'mathematical' ideas and
concepts in our ontologies that they will thereby come to be somehow
more rigorous and powerful than if we use 'nonmathematical' concepts,
on the grounds that mathematics is kind of tougher than mere logic.
This is just not the case. It is perfectly fine to use such concepts,
but the only content they provide is whatever content can be
axiomatized using them, just like any other formally axiomatized
concept. They certainly do not 'come for free' in any sense: writing
(RealNumber ?x) does not make ?x into a real number any more than the
earlier sentence made it into a physical object. Which is not to say
that we cannot take advantage of the very large amount of such work
that has been done by professional mathematicians, of course: only
that there isn't anything magical about the math-ness of the concept
names. All we ever get, in the end, are axioms, and if some intent
isn't captured in the axioms (or in some other special-purpose
reasoning code) it's not in the ontology.
Pat Hayes
---------------------------------------------------------------------
IHMC (850)434 8903 home
40 South Alcaniz St. (850)202 4416 office
Pensacola, FL 32501 (850)202 4440 fax
phayes@ai.uwf.edu
http://www.coginst.uwf.edu/~phayes