Date: Sun, 6 Oct 91 10:29:54 EDT
To: GIO@earth.stanford.edu, NECHES@isi.edu, SRKB@isi.edu, INTERLINGUA@isi.edu,
KR-ADVISORY@isi.edu, GINSBERG@t.stanford.edu, SKPEREZ@mcimail.com
Subject: Multiple ontologies
Gio and Bob,
I'm glad that everybody is thinking about multiple ontologies, but I
don't believe that we have to solve all the problems concerning them
all at once.
The important thing in the standard is to accommodate whatever kinds
of reasoning systems people want to implement. There have been various
research prototypes described in the AI literature that reason about
different participants' goals and motives from different points of view.
The standard only has to provide a framework that can accommodate such
uses -- it doesn't have to invent all possible uses in advance. Then
if some clever person wants to design a God module that can peer into
the hearts and minds of the other modules and reason about them, that
should be a possible way of using the framework.
Doug Lenat gave an example that illustrates a practical need for
different levels: A physician, for example, would have a detailed
ontology of medical terms, including a clear distinction between symptoms
and diseases. But a typical patient is likely to confuse the distinction
and say "I suffered from a cold" or "I suffered from a cough." When
the doctor module talks to the patient module, the doctor has to
reclassify the input according to a different ontology.
Re people who confuse distinctions: I was talking with a knowledge
engineer who was designing a diagnostic system. He told the domain
expert, "I'm not sure what you want the system to do -- determine
that a problem has occurred, determine the cause of the problem, or
suggest a way of fixing the problem." Whereupon the "expert" asked
"What's the difference?"
When you have interacting modules, you have to deal with conversations
like that. The standards don't have to solve all the problems, but
they should allow for new ideas that might solve some of them.