Reply to Matt G.
sowa@watson.ibm.com
Message-id: <9202040229.AA15046@cs.umn.edu>
Reply-To: cg@cs.umn.edu
Date: Mon, 3 Feb 92 21:29:12 EST
From: sowa@watson.ibm.com
To: ginsberg@t.stanford.edu
Cc: cg@cs.umn.edu, interlingua@isi.edu, kr-advisory@isi.edu, srkb@isi.edu
Subject: Reply to Matt G.
Matt,
Some comments on your comments on my comments:
>> Until we can agree on what
>> these proposed "standards" are supposed to standardize, we cannot even
>> begin to agree on what the languages should look like.
>
> That is certainly at odds with the discussions that have been taking
> place thus far!
In several of my notes during the past few months, I have repeatedly
stated that we should have a clear list of requirements. But nobody
picked up on that point. The only word that seems to stir any passion
is "standards". After all the email flak I've been getting today, I'm
beginning to regret the fact that I used it in my note yesterday. I can
really sympathize with Bob Neches standing in front of the gang at KR'91.
>> I do believe that if you had a
>> complete description of a Boeing 777 in FOL, you could do all sorts
>> of creative reasoning with it using your super nonmonotonic reasoner.
>> You could start with the description as a collection of low-level facts,
>> and do hypothetical, nonmonotonic, or fuzzy reasoning about what would
>> happen if someone set off a bomb in the lavatory.
>
> The whole lesson of nonmonotonic reasoning is that this is exactly
> what you *can't* do. It just isn't the case that you can apply
> nonmonotonic, or hypothetical, or fuzzy reasoning to an FOL database.
> All of these methods rely on information that an FOL database is
> simply lacking. What predicates do you minimize? The FOL description
> has doubtless omitted a lot of abnormality conditions because the
> designer wasn't interested in bombs in toilets. (I've seen parts of
> the Rockwell description for one of their planes; it certainly had
> this property.)
True. If all you have is a list of FOL statements, there is no place
to begin doing any nonmonotonic reasoning. However, such a description
would be an excellent starting place to which you could add various
nonmonotonic rules.
The point I was making is that a statement of what exists is a necessary,
but not sufficient prerequisite for any kind of hypothetical or
nonmonotonic reasoning. Having such a description would be a big help,
even if it didn't give you all you need.
>> A standard must be able
>> to communicate any statement of fact about anything in the universe.
>> That goal can be achieved with nothing more than first-order logic.
>
> Or LISP, for that matter. Why not just exchange code? The reason, of
> course, is that we want a language that is perspicuous, not just any
> language at all.
True. Aren't you agreeing with me on that point?
>> I also believe that if anybody were able to develop
>> a super nonmonotonic Sergeant-Friday-simulator, they could turn it loose
>> on "just the facts" and do all sorts of "wild and wonderful things."
>
> The fact is, they couldn't.
Sergeant Friday had a lot of nonmonotonic hypotheses in his head, which
he acquired by long experience. But in any particular case, all he was
looking for from interviewees were the facts. Their guesses about what
those facts might mean were probably so far off the mark that they would
not be of any help to him.
In the same way, a system that could at least communicate a set of facts
in FOL would be very useful as a source of evidence for a nonmonotonic
reasoner. But that reasoner would also have to add further premises from
its own knowledge base.
> Let me propose -- again -- that the standards folks take a look at my
> proposed language. It subscribes to the *syntax* of FOL, but not to
> the semantics.
What proposal was that? If you mean the one in the Fall '91 AI Magazine,
I said before that I liked it and that we should seriously consider it.
But in any case, you just used the word "standards". I thought that term
was anathema to you.
John Sowa