COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	1. More typically declarative information that isn't close to imperative.
C00004 00003	Jan 18
C00009 ENDMK
C⊗;
1. More typically declarative information that isn't close to imperative.
A cat weighs from six to 20 pounds
No breed of dog has blue hair
Rosemead is a city near L.A.
Jimmy Carter's nephew just got out of jail.

2. Is formulation 6 on p. 5-3 available to the mechanic's wife for
explaining the failure of her son to start the car?

3. C. proposes putting the information into a production system.  Is
a production system adequate, or can it be adequate, for dealing with
unknown preconditions.  fFor example, the mechanic's wife attempts to
floor the accelerator, fails, and discovers that some joker has wired
the accelerator to the dashboard.  Of course, she thinks of removing
the wire, but would the production system on noticing the faailure
of a production, be able to do this.

4. Creative generalization needs to be distinguished from the
simpler formal generalization.

5. I am dubious about "co-operative".
Also about experts more general than subroutines.
Jan 18

	1. The Huberman chess end game program provides an example
of generalized planning.  Namely better(position1,position2) can
be regarded as a plan for the mate.  It generalizes the conventional
notion of planning in AI, because the number of stages of the plan
is not specificied.  It says search for a way to force an improved
position, and this can be regarded as a subgoal, but the number of
subgoals that are required to reach the final goal is not determined
by the planner.

	A further point is that the better(p1,p2) need not be compiled
into the program, although it can be regarded as reducing the main
problem to a sequence of well-defined sub-problems.

	It might be worthwhile to look for an example where the heuristics
consist of a criterion for progress in a non-game example.  Experiments
could then quantify the effect of improved progress heuristics in
reducing combinatorial explosion.  The other side of the Huberman
heuristics, namely recognizing moves that make things worse should also
be used.  Slogan - "Planning reduces problem-solving to hill-climbing".

	2. I would prefer to soft pedal (i) compiling and (ii) information
retrieval in order to get some clearer examples of the direct use
of declarative knowledge.

	3. Maybe there is a common generalization of Huberman
planning and GPS planning.

	4. Slogan - "Design precedes construction planning".
tower example, chess example.  Much AI problem solving ignores
design.  Design is a form of planning.

Creary has two more points:

	Why isn't there more interaction with HPP (also with
philosophy, psychology, etc.)?

	What bothers him about Doyle talk.

1. lacks utility theory.  jmc - good

jan 25

Do we need examples of non-physical design, e.g. of an organization?

In describing action rules, we prefer a level of detail at which
actions have definite effects.  In general, goals are given at
a lesser level of detail than is required for determining the
effects of actions.  Therefore, we need to distinguish between
"states of affairs", which are at that level of detail, and
propositional fluents which may be at a lesser level of detail.

More generally yet, our knowledge of the effects of actions may
be at a lower level of detail than is required to determine
the preconditions for the desired subsequent actions.  This forces
plans to contain servos, i.e. while loops, and conditional
branches.  Example: if we have many heavy boxes to move, our
plan envisages resting whenever the moves have made us tired,
but our knowledge of the effects of moving boxes is insufficient
to determine when we will be tired.  Thus using incomplete states of
affairs is mitigated by including servos in the plans.

C. wants to make a 10 year plan but would start with a travel planner
with a later added advice-taker component.  He will look at the Gus
paper in the AI Journal.  What about doing the epistemology first?

Creary has a "time driven agenda system for an executive for a
general intelligence", but it isn't ready to be read yet.
DGI.TXT[DGI,LGC].  Access dgi,lgc.  password on paper.