[mod.ai] Seminar - Granularity

LANSKY@SRI-AI.ARPA (05/08/86)

			      GRANULARITY

                    Jerry R. Hobbs (HOBBS@SRI-AI)

          Artificial Intelligence Center, SRI International
                         CSLI, Stanford University 

   	 	        11:00 AM, MONDAY, May 12
         SRI International, Building E, Room EJ228 (new conference room)

We look at the world under various grain sizes and abstract from it only
those things that serve our present interests.  We can view a road,for
example, as a line, a surface, or a volume.  Such abstractions enable us
to reason about situations without getting lost in irrelevant
complexities.  Knowledge-rich intelligent systems will have to have
similar capabilities.  In this talk I will present a framework in which
we can understand such systems.  In this framework, a knowledge base
consists of a global theory together with a large number of relatively
simple, idealized, grain-dependent local theories, interrelated by
articulation axioms.  In a complex situation, the crucial features are
abstracted from the environment, determining a granularity, and the
corresponding local theory is selected.  This is the only computation
done in the global theory.  The local theory is then applied in the bulk
of the problem-solving process.  When shifts in perspective are
required, articulation axioms are used to translate the problem and
partial results from one local theory to another.  In terms of this
framework, I will discuss idealization, the concepts of supervenience
and reducibility, prototype-deformation types of description, and the
emergence of global properties from local phenomena, and the
relationship of granularity to circumscription.  Several examples of
uses of this framework from a wide variety of applications will be
given.


VISITORS:  Please arrive 5 minutes early so that you can be escorted up
from the E-building receptionist's desk.  Thanks!