THE BIG PICTURE ***READ THIS!***

Doug Lenat <lenat@cyc.com>
Date: Tue, 6 Jun 1995 15:25-0500
From: Doug Lenat <lenat@cyc.com>
Subject: THE BIG PICTURE ***READ THIS!***
To: fritz@rodin.wustl.edu, doug@csi.uottawa.ca
cc: cg@cs.umn.edu, srkb@cs.umbc.edu
In-reply-to: <9506061754.AA05265@rodin.wustl.edu>
Message-id: <19950606202539.9.DOUG@ETHEL.MCC.COM>
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk

Fritz,

You have hit the nail on the head, here, I'm afraid.  I will tell you
why we (the Cyc group) don't write more such papers and come to more
workshops and for that matter even respond to more notes on SRKB.  I
will close by touching on KIF, and what's wrong with this whole field,
thereby probably alienating many folks on this list, but maybe also
doing some good.  So read on.

When we made a big effort to have a high profile and actively
participate -- back at the first workshop that McCarthy/Morgenstern/...
et al arranged on Formalizing Common Sense, a few years ago at a
Stanford AAAI Symposium -- we had 6 Cyc people submit 4 papers (I
believe the numbers are right, but I may be off by +/- 1), all of which
got accepted, and 3 of which were on specific ontologies (one on
buying/selling/money, one on the space of information-bearing objects,
and I think the third was on daily human activities.)

They were all accepted, and everyone prepared their talks, but we
discovered to our chagrin when we arrived that the 3 ontology papers
were all squeezed into one single timeslot!  When we asked (okay,
complained) about this, we were told that surely no one was REALLY
interested in hearing about the finished ontologies, or even the
experiences and lessons learned in actually codifying and testing them;
no, we were told, "real scientists" don't care about artifacts like
that, they want to talk ABOUT the theoretical philosophical/logical
problems in building/sharing...  ontologies, things at the level of the
Yale Shooting Problem, and when to treat extensional failure as
intensional negation, etc.

To top it off, this session was scheduled at the very same time as the
4th paper was to be delivered, in a session called something like
"Implemented Systems" -- the irony being that the organizers felt that
surely no single individual would be interested in both "formalizing
common sense" and "real, implemented systems."

So we've kept a pretty low profile ever since, considering what we've
been doing and what we've produced.  I'm gratified to see notes like
yours, which suggest that the field is maturing, by which of course I
mean "seeing things more my way." ;)

Since I've clearly spent most of my diplomacy-points here for a while
anyhow, I'll close by spending the rest, and obliquely address your
comment on KIF.  KIF is a good example of what's wrong with this field
-- no, not the language itself, the way that it is being developed and
what is/isn't expected from it if only it were perfect.

Guha wrote a KIF-to-CycL translator in a weekend -- that's about ten
days in human time, given how fast Guha programs -- and I'd be very
interested if there are any new features in KIF which aren't in CycL
(and weren't, all along -- the last big extension that was made to it
was adding contexts, 4 years ago.)  Conversely, CycL grew from a vanilla
frame-based language (in 1984), with new features being added when --but
only when-- they were VERY necessary, to represent huge numbers of
commonsense assertions and, less frequently, application-specific
knowledge.

Thus, of course we've allowed second-order relations (since 1985) and
even limited (stylized) nth-order quantification (since about 1989), not
because we wanted to but because we had to, to enable, respectively,
most meta-level things (preferences, arguments, history, etc.) and
reasoning in predicate-space.

To the extent that CycL is probably therefore necessary and sufficient,
and to the extent that it is 98% FOL, we find it constantly amazing that
folks keep groping toward something that's "their own"; that folks are
doing it committee-wise; that folks are driven by discussion more than
by plethorae of examples; and finally that folks believe that there's
much power of ANY sort to be derived from having a common syntax such as
KIF or CycL or FOL.  The CYC member companies okayed our release of CycL
and most of the Cyc ontology (including everything that philosophers
mean by the term), and several groups have partaken of them, but
surprisingly few.  But this is not my point -- let me get back to it.

As we learned from our Frames days, having a deficient repr. language
can hold you back (if you can't easily represent disjunction, negation,
modals, higher-arity predicates, etc.) but having a good one, no matter
how wonderful it is, is just the tiniest tip of the iceberg, the glimmer
of the first prerequisite for knowledge-sharing.  

In certain specific cases of problem solving, having a good repr. can
make a big difference (cf. Saul Amarel's classic work on the
Missionaries and Cannibals problem); but most of the time we as human
beings share knowledge using much less specialized reprs. such as
english.  Believe me, FOL -- with a few second-order extensions -- is
adequate.  CycL is adequate.  By now KIF is surely fine, too.  So my
advice to all of you is: stop tinkering with KIF, start sharing your
ontologies, and you will get hit in the face with what's really
important, namely sharing most but not all of the meaning of most but
not all of the terms most but not all of the time.  To those of you
who've already gone far along that path, I apologize for being so
patronizing, but for the rest of you:  get over your physics envy,
or philosophy envy, or whatever, and get down to the real work!

--Doug Lenat