Re: SUO: Re: Ballot Comment
John,
Maybe there is another way to look at this which is compatible with both
our views. I'd suggest that in practice, the notion of a lattice of
theories can be implemented with simple inheritance and default reasoning
(to the extent that direct contradictions are not needed). Maybe it's that
I've been envisioning a system of default reasoning operating on the
ontology and possibly you haven't that is causing the apparent difference
of opinion.
I was also not envisioning that the SUO would require anyone to convert
their internal representation to SUO. That indeed would be an untenable
position. The SUO will, I believe, be suitable as an interlingua for
existing systems as well as a basis for new systems. This is also distinct
from suggesting that SUO terminology be adopted at the user interface level
in different domains which I also think would be inappropriate.
As an aside, I know that you and McCarthy as well as others have written
about contexts (although not always by that name) prior to Cyc. My point
was just that Cyc itself isn't an example that maintaining different
contradictory theories is required since they don't actually do that in
practice.
I'm impressed by Robert's work but don't yet see any advantage to
merging SUMO with IFF. If someone else does, they should begin that merge
and show us why it is of value. A number of folks have made the quite
legitimate point that a case needs to be made as to why category theory
buys us anything over set theory. This would have to be in the form of
concrete examples, with axioms, because intuitions just aren't enough.
Adam
At 09:55 AM 8/17/2001 -0400, John F. Sowa wrote:
>Adam,
>
>I'm not basing the idea of multiple theories on Cyc. I was recommending
>it before Cyc even adopted that idea. (Guha cited my 1984 book in his
>1991 dissertation as one example of the multiple context idea.) The
>notion of the infinite lattice of all possible theories can never be
>implemented as such (since it's infinite), but the idea is an
>abstraction
>of a large number of implemented systems. Every belief revision system
>can be classified as such, and every nonmonotonic logic or default-based
>system of any kind can be classified as an implementation of such a
>lattice. And as I said in my talk, it fits very well with
>Wittgenstein's
>theory of language games, among others.
>
>Furthermore, any monolithic ontology that forces the world to convert
>to its terms is paddling against the tide. There are literally millions
>of databases and kn. bases in the world that have their own peculiar
>ontologies, and the only way to accommodate them is to give them a
>migration path from where they are today to where they are more likely
>to find a workable niche in the future. To some extent, Cyc does that
>with their microtheory notion, which allows legacy systems to coexist
>with the Cyc KB.
>
>Some comments:
>
> > That's a fine goal in theory but if you or others feel that's the right
> > approach then I think you have some responsibility to get specific about
> > exactly what terms and axioms from SUMO should be related in what way to
> > particular terms in IFF.
>
>Indeed, somebody has to do that. IFF makes room for multiple theories,
>and SUMO was formed by putting together multiple theories. So one place
>to start is to look at the original sources and determine whether that
>breakdown is appropriate. The work that went into SUMO has already
>made the theories more compatible than they were to start with, so I
>would recommend that the SUMO axioms be subdivided into theories that
>are the same or similar (allowing for whatever improvements were made
>in adapting them to SUMO).
>
> > Well, then I'm confused by your point. Either we start with a blank sheet
> > of paper and work on those fundamental distinctions, or we critique an
> > existing document. Which is it that you prefer? Or do you advocate a
> > third option?
>
>I thought that SUMO was making good progress as a small project, but I
>also thought it was premature to suggest it as a candidate for the
>ultimate SUO product. I believe that more work was needed, especially
>in addressing the issues of interoperability with legacy systems.
>The lattice of all possible theories makes room for anything, and it
>shows how to move from one theory (such as a legacy system) to another
>(such as SUMO or some subset or version of it) by the basic operations
>of belief revision.
>
> > >I said that the way SUMO is going is not the way I would recommend.
> > >But I have suggested an alternative, which is to combine it with the
> > >IFF approach.
> >
> > ah, but how...
>
>As I said above, get Ian and Robert together to see how they can
>partition the SUMO axioms into Robert's pigeonholes.
>
>John
Adam Pease
Teknowledge
(650) 424-0500 x571