COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	frame[f86,jmc]		The Frame Problem Today
C00006 00003	"csdept@ukans"@csnet-relay
C00007 ENDMK
C⊗;
frame[f86,jmc]		The Frame Problem Today

For Frank Brown's Workshop on Logical Solutions to the Frame Problem

extended abstract due Dec 1
csnet csdept@ukans

	The frame problem was first posed in (McCarthy and Hayes 1969).
It is the problem of avoiding having to state what doesn't change in giving
rules describing the effects of actions.
For example, if we have facts about moving blocks and
painting blocks, it is normally true that moving a block doesn't change
its color or that of any other block or the locations of other blocks, and
painting a block doesn't change the positions of any blocks and the colors
of blocks other than those painted.
When there are only a few actions and properties of objects it is possible
to give the required {\it frame axioms} explicitly.
Here they are for moving and painting blocks using the situation calculus.
In other formalisms the rules can be expressed similarly.

	However, if we
are building a general purpose database of common sense knowledge,
there will be some large number $m$ of kinds of action and another
large number $n$ of properties of objects, and it seems unreasonable
and even infeasible to write down the $mn$ axioms that would be required.

	Moreover, human performance tells us that this must be unnecessary,
because we aren't conscious of large numbers of frame facts, and when
we describe the effects of actions to other people we don't include
large numbers of such facts in our oral or written discourses.

	The name {\it frame problem} came from one approach to the
problem based on the idea of {\it co-ordinate frame} used in
mathematical treatments of geometry.  It is also analogous to the
concept of {\it assignment statement} used in computer programming.
If we attach the properties of objects to ``frames'' we can define
actions relative to a frame by asserting that no other properties
attached to the frame change besides the ones listed.
  More recently some
people have referred to the ``persistence problem'', and maybe this
would have been a better name.
  Here is an example
of such axioms for moving and coloring blocks.

	While using frames reduces the number of axioms to a number
of actions, it still seems more rigid a notation than is plausible
as a model of human knowledge or likely to be satisfactory for an
evolving common sense database.
"csdept@ukans"@csnet-relay
for Frank Brown
I plan to submit an abstract for the frame problem conference, but
I would like till next Monday if possible.  The title will the
"The Frame Problem Today", and it will contain both historical
treatments and my current candidate using circumscription.
			John McCarthy