Re: SUO: Program Semantics
Bill,
In your list of options, you forgot the biggest one: Cyc has been
pursuing the "pragmatic" approach for 18 years (it started in 1984).
Doug Lenat has been proudly proclaiming that he never put any feature
into Cyc unless they found some application that needed it.
By 1991, Lenat had discovered a need for a solid logic-based approach
instead of frames, he had dropped the certainty factors of his "fuzzy"
approach, and he had added contexts and microtheories. Those were all
recommendations that I had made in my 1984 book (actually published at
the end of 1983).
I'm not claiming that I was using a crystal ball or Tarot cards.
What I did was to analyze the published literature of AI, study
what worked and what didn't, study the theoretical foundations for
the various approaches, and make an informed estimate of what should
be done. Doug could have saved many years of effort if he had
taken the advice in my book when he was starting his project.
What I am now recommending for SUO is based on another 18 years
of the same kind of study of both theory and practice. And Cyc is
one of the important projects that I've been watching over the years.
My major complaint about your option A (adopted by SUMO) is that it
is a warmed-over attempt to redo what Cyc has already invested 500
person years in doing without coming close to their goals for 1994.
Some comments on your comments:
BA> 1) A successful search for "a proper structure for doing ontology"
> depends on achieving some agreement on what ontology is. I just
> don't see that coming along any time soon. I see at least three
> camps here:
I think that we are very close to agreement. The major obstacle
now is the SUMO group, which has been ignoring the evidence.
BA> Camp A: We'll call this Microsoft camp for convenience. Believes
> that having a standard stock of categories of sufficiently broad
> reach and functionality that users will be attracted to use it,
> thereby achieving interoperability more or less by fiat. This
> approach is completely pragmatic, laying no claim to, but willing
> to borrow from theory in whatever field may have something to
> offer.
Your name is inappropriate, since Microsoft right now is doing
something quite different. Bill Gates had been a sponsor of the
Cyc project in the early '90s, but he dropped the sponsorship partly
as a result of some influence from my former buddies at IBM -- George
Heidorn, Karen Jensen, and Steve Richardson. George and Karen have
recently retired, but Steve is pursuing an approach they had initiated
at IBM: parse machine readable dictionaries and other texts to build
up a knowledge base. Their current MindNet system doesn't have axioms,
but it is a very solid (and much larger) competitor to WordNet. It is
not theory based, but it's good as far as it goes.
SUMO might be a better name for Camp A, except for the last line of
your description. So far, I haven't seen any evidence that the SUMO
group is willing to "borrow from theory".
BA> Camp B: We'll call this the Peirce camp for convenience. Believes
> that those in Camp A are nuts - that only some sufficiently subtle
> method of reconciling language use amongst users/systems will work.
> This view is strongly connected with an anti-realist philosophical
> view that there is nothing "out there" in the world to get hold of
> as a basis for doing ontology.
Again, your choice of name is wrong. Peirce was very much a realist,
and one of the leading experimental (and theoretical) physicists of
his day. Besides his many achievements in logic, Peirce was the first
person to recommend the use of a wavelength of light as a standard for
defining the meter. And then he designed and built the equipment to
use light waves to measure the pendulums he had designed for measuring
gravity. So I have no idea which SUO subscribers belong to Camp B.
BA> Camp C: We'll call this the the formalist group for convenience.
> Believes that some sufficiently general mathematical principles can
> be applied to enable interoperable domain ontologies to be built,
> but posits only a minimal set of categories needed to make the
> system go, but that these are not strongly connected to any signs,
> images, symbols, words, or anti-realist concepts.
Again, I don't know who you are talking about here. The IFF group is
certainly formalist, but they have been developing a framework that can
accommodate an infinite set of categories. They are happy to accept
SUMO, OpenCyc, WordNet, IMPS, and any other set of categories anyone
can give them. As far as I can tell, they are the most pragmatic
of all the SUO subscribers.
I'll skip over several paragraphs of your note, since you don't mention
any names, and I have no idea whose positions you are criticizing. But
you do mention one group that is familiar:
BA> The WWW community has taken this view to the limit, proposing a
> dizzying array of semantically crippled quick-fix languages and
> claiming to be doing "ontology".
I certainly agree. They have ignored everything that has been done
in AI for the past 40 years. The RDF notation, which was originally
designed by R. V. Guha, who was formerly the associate director of Cyc,
is a trivial subset of what Cyc was doing in 1984. Guha proposed the
version 1.0 of RDF, and he hoped that a much richer language would be
developed for version 1.1. But since then, the W3C has been hung up
on pursuing such low-level details that Guha withdrew in disgust. He
is still on the RDF committee list, but he hasn't participated in any
of their work for over two years.
BA> In my view we can do much better without crossing the line into
> "big question" territory.
I agree that we can do much better, but I must point out that the W3C
has never even approached "big question" territory. They are still
mucking around in extremely low-level details.
And I'll comment on the following point that also mentions some names:
BA> I think our first order of business should have been to entertain
> a couple or three truly "top-level" structures. IFF represents one.
> Nicola's stuff represents another. And perhaps the Peirce camp could
> offer a third. Then we run all three off against a series of mutually
> agreed upon challenge problems and let the chips fall where they may.
I see these proposals as mutually supportive, not competing. I believe
we should synthesize them into a framework that includes all of them.
And I've done that in my paper, "Signs, Processes, and Language Games":
http://www.jfsowa.com/pubs/signproc.htm
One paragraph I completely agree with is your conclusion:
BA> If only 50% of the effort spent arguing on this list were directed
> at some effort(s) to build something which can be evaluated by someone
> without a PhD in philosophy or computer science, then there would be
> no need to worry about things being forced down our throats - the
> results would speak for themselves.
Amen, I say. And to back it up, I've been working with a group of
people who have formed a new start-up company called VivoMind LLC,
which is doing exactly that. For a brief overview of the kind of
technology that is going into the VivoMind company, see the slides
I presented at the Knowledge Technology 2002 conference on March 13th:
http://www.jfsowa.com/talks/negotiat.htm
If you want to see how to map different ontologies that describe the
same structure, read the slides starting with the title "Representing
a Physical Structure" and ending with "Mappings found by VivoMind".
If you want to see how the technology scales up to real applications,
read the slides starting at "VivoMind for Legacy Re-engineering".
That is an implemented application that saved several million dollars
for the company that had employed Majumdar and Leclerc to do the work.
If you want an overview of the kind of technology being developed by
VivoMind LLC, see my paper "Architectures for Intelligent Systems":
http://www.jfsowa.com/pubs/arch.htm
If you want to know some of the people involved in the company, read
the names mentioned in the slide "Implementing the Answer".
John Sowa