KR standards and shared KB workshopHayes@mcc.com
Date: Thu, 21 Mar 91 15:37 PST
Subject: KR standards and shared KB workshop
Cc: firstname.lastname@example.org, SRKB@venera.isi.edu
Tom, reply in the form of interjections to your message ( which I had
not received earlier, by the way )
you are interested, you should come to the workshop on March 24-25 at
Pajaro Dunes, where the shared KB group meeting. Since that is so
soon, we need to know whether you can make it. (Doug was invited a
long time ago, but is not sure whether he can make it. You could take
his space with no impact on the housing reservations; you're
welcome to come regardless - just let us know.)
OK, I will be there on Sunday. But I will drive back Sunday evening and
may not make it on Monday ( It depends on whether or not I have to go to
ANOTHER meeting ), but if I do then I will drive down. Ie, no bed needed.
I would like to talk with you on the phone about this,
Try dialling 323 6797 next time...
but in the mean
time here are some relevant documents to set the background.
Please address some of the following questions:
* What knowledge would be useful
To whom?? What is the intended community of sharers?
* What are the barriers to knowledge sharing? How can AI help?
What? I thought this was already an AI project. This question suggests
that knowledge sharing is a pre-existing problem area. Is this true, and
if so ( to repeat ) , amongst whom?
* What should be the unifying vision for knowledge sharing and
reuse? (e.g., Stefik "The Next Knowledge Medium" (AI Magazine) or
Lenat and Feigenbaum "On the Thresholds of Knowledge" (AAAI and AI
There shouldnt be a unifying vision. Thats not how good research is
done. People need to be able to pursue their own visions.
* What are the critical research issues that need to get solved?
Rather a lot, unfortunately. Like, for example, solving the knowledge
Initial Working Groups
On Sunday we will divide into three working groups:
* WG1: Sharing domain-specific knowledge (engineering,
manufacturing, logistics, medicine, biology)
But most of this hasnt even been formulated yet, so how can we decide
how to share it? Formalising engineering and manufacturing and biology
are new research domains. Example: should engineering concepts have
temporal extents? Another: in describing hardening processes should I
introduce surfaces as a separate ontological category ( to talk eg of
surface hardening ), or not? If I do, do 3-d objects touch or do their
surfaces touch? Just about everywhere one looks there are questions like
these, and we don't have clear answers to any of them: or even can be
really sure that we are asking the right questions exactly. What kind
of geometric structure shall I take space to have, euclidean,
qualitiative, tolerance-based, or what? ( And if Im a young researcher
with some wild but potentially valuable idea, will I feel pressure to
not work on it because it wouldnt be compatible with the Standard? )
* WG2: Sharing domain independent knowledge (natural language
processing, common sense, reusable software, integration of databases,
semistructured knowledge / hypermedia)
My above remarks are even more true here. Im sorry, but I have to say
what I said to Mike G. the other day: to suggest that there should be a
Standard ontology for natural language and common sense is nothing short
of irresponsible for anyone competent in AI. Cyc is the only attempt to
even try this idea out thoroughly, and it is discovering all sorts of
difficulties and complexities, and doesnt even claim to have got it
right or even to know for sure whether it can really be done: its an
experiment. And you want to set a Standard?? Fortunately, the idea is
so ridiculous that it hasn't a hope in hell of getting anywhere. I am
more worried about WG1, which could easily settle on some kind of
industrial 'standard' and seriously damage progress in these difficult
and potentially very rewarding areas.
* What knowledge is useful to share?
Again, useful to whom?
* What are the barriers to knowledge sharing?
The worst one is mutual incomprehension. Which is exactly our problem:
I dont think AI has much to offer on this dimension.
Look, Im all in favour of creating large Kbases, and I see the
desirability of making existing ones accessible for use by others.
Maybe we should think of ways of doing this, including such things as
what might be called ontological documentation so that people can more
easily write translators from one KB to another language. But this is
not the same as suggesting the use of a single common language or, still
less, a common ontology. That is a much more radical suggestion, rather
like a student of Newton's suggesting that all mathematics from then on
be done in terms of fluents so that people can understand one another
better. We have lived through such attempts to standardise a programming
language - COBOL was one ( and it was successful, in much the corrosive
way I am afraid of here ), ADA another, and so on. But a standard
ontology is much more like suggesting having standard programs, not a
standard programming language. And that analogy should illustrate how
braindamaged the idea seems to me to be.
Am I completely off your wavelength, or do we really disagree this much?
Anyway, I will be wellbehaved on Sunday: my private goal is to find out
>From the horses' mouths what people think about this stuff. Who knows,
maybe I can be converted.