Re: Report on ANSI X3H4 meeting

rjb@research.att.com (Ron Brachman)
Message-id: <m0lB94m-000suhC@rocket.research.att.com>
Reply-To: cg@cs.umn.edu
Date: Mon, 3 Feb 92 14:11 EST
From: rjb@research.att.com (Ron Brachman)
To: pkarp@ai.sri.com
Cc: sowa@watson.ibm.com, kr-advisory@ISI.EDU, srkb@ISI.EDU,
        interlingua@ISI.EDU, cg@cs.umn.edu, mzemanko@note2.nsf.gov,
        jwinkler@nasamail.nasa.gov, skperez@mcimail.com,
        jfulton@atc.boeing.com, mjones@atc.boeing.com
Subject: Re: Report on ANSI X3H4 meeting

Nicely put, Peter.  One crucial aspect of your statement is central to
my (and many others') concern with the direction in which things seem
to be proceeding, and since you have crystallized so well an issue
that has been simmering for a long time, I thought I would jot down a
few comments.  Here's your key phrase:

	We should expect that only the best understood
	aspects of knowledge representation theory have been standardized, 

So much of the discussion on the various knowledge-sharing and
"standards" lists in recent months has been focused on speculation,
research, and the design by committee of the
knowledge-representation-to-end-all-knowledge-representations, it is
downright scary.  Quite some time ago, Matt Ginsberg made a proposal
for a small, simple, core of representation technology that might
serve as an interlingua; the idea was not taken very seriously by the
Interlingua committee for various reasons, and since then, the
discussion has gotten farther and farther afield of "the best
understood aspects" of KR.  People seem to react defensively and
angrily if the proposed language doesn't cover some aspect of their
own system, almost as if they were saying, "Well, if it doesn't handle
xxxxx, which is needed by some systems for natural language processing
(or whatever), then it isn't adequate and I sure won't use it...."
This is what is killing this effort and dividing the field.  Various
people have argued reasonably for the value of standards, and there
are indeed situations in which standard languages for communicating
(parts of) knowledge bases will soon be useful and necessary.  But
what is happening is the attempt to do research over the Internet and
design the world's most complex language based on constructs and ideas
that have not been shaken down on real problems.

In fact, another reason that the effort seems to be going haywire is
that it is not being driven by a specific real need.  If DARPA would
provide a set of REAL knowledge bases that need to be shipped from one
place to another, then we could evaluate proposals based on whether or
not they address questions that really arise in this scenario.
Without this realistic forcing function, most argument is idle and
academic.

Bottom line: if we are going to develop some standards, just as you
say, they should be based on well-understood constructs that have held
up under fire in realistic settings, and we should not yield to the
temptation to develop a KR formalism that subsumes all others,
especially by an ill-defined loosely coupled group of people who want to
do research by e-mail.  Many of us are feeling threatened, angry, and
disappointed at the current effort because it is not focusing on
things that work and are understood.  By my lights, the KRSS part of
the effort IS an attempt to tackle a small set of things that has
worked well in practice, is widely used, and (we think) except for
syntactic variations, is pretty much already consensually agreed on.
Unfortunately, that effort may also be threatened - I have heard on
occasion that KRSS won't "work" because it doesn't account for some
constructs in some (very expressive) languages.  I hope it won't start
down that slippery slope the way some other efforts have.

We can all of us together develop extraordinarily powerful languages
if we want, but it is ridiculous to support any such RESEARCH effort
as a "standard" until it was thoroughly debugged and found workable
in practice.  

All of this was part and parcel of the major community-wide discussion
on this topic that was held at KR'91, and at which many of us thought
the issue was settled and that the steering committee for the
knowledge-sharing effort was supportively in tune with.  Unfortunately, if
the e-mail traffic over the last few months is any indication, the
clear message of that meeting has once again been lost.  Part of the
confusion has clearly been engendered by the injection of the PDES and
ANSI discussions into the DARPA discussions, and by people assuming
that the DARPA effort was a standards effort similar to those efforts.
Unless there has been another reversal since KR'91, that is NOT the
key thrust of the knowledge-sharing research effort, and I for one
wish that we could clarify things and get back on the right track.  

Ron