COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	query[f83,jmc]		Research queries
C00007 ENDMK
C⊗;
query[f83,jmc]		Research queries

Is circumscription good enough to express the non-monotonic presumptions
we want to put in axiomatizations of common sense?

The object is to minimize abnormality and also to finitize.
The wolf, goat and cabbage problem (WGC) is better than MC for a first
example of finitizaton, because we don't have to represent the
symmetry.  The object of the finitization is to get a system with
a finite number of states to which exhaustive methods can be applied.
We want the conclusion that the problem is unsolvable if the boat
can carry only one or if the wolf eats cabbages.

Suppose we formalize the common sense knowledge relevant to WGC.
Our objective will be to go from the formalization to the 16 state
model.  It would be most impressive to make separate actions out of
putting things in the boat, getting in the boat, rowing across and
removing things from the boat.  Then the actions of the model will
have to be compound actions in the world if the model is to have just
the 16 states.  We should allow killing the animals and eating the
cabbage as actions.  They will get left out of the model, because
they are seen to preclude solving the problem - likewise overloading
the boat.  The redundant actions of putting something in the boat
and then immediately taking it out again should be possible.

However, the problem will be easier if we make transporting something
across the river by boat an elementary action.  The problem of setting
up the model and circumscribing the facts to validate it seems quite
hard as it is.

There seems to be some kind of generalized definition involved.  Let's
discuss with Jussi whether a generalization is required of the
definition features of EKL.  Also putting circumscription in EKL.

The circumscription involves the set of actions known to be relevant
and not known to be harmful.  We have some positive reasons to regard
some actions or sequences of actions as relevant and we jump to the
conclusion that these are the only relevant actions.  Alternatively,
we may circumscribe on certain relevant properties of the state, e.g.
on which side of the river the various objects are.

When someone says "The only relevant features of the situation are
which bank are the farmer, wolf, goat and cabbage", at what level
of metalanguage or reification is he working at?  Is this a sophisticated
usage or would children understand it and even generate something
equivalent?  It would seem that bright children at least must be
pretty good at discovering the relevant aspects of situations.