COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002		The essential innovations of Doyle's "deliberation and decision
C00007 ENDMK
C⊗;
	The essential innovations of Doyle's "deliberation and decision
system did not seem to be clear from his lecture.  
Here is a slight modification of his example that seems to me to
illustrate one of the essential ideas:

	We have the rules

chivalry:	approaches(door,x,s) ∧ woman(x) ⊃ should(open(door),s)

lib:		woman(x) ∧ libber(x) ⊃ inapplicable("chivalry")

courtesy:	approaches(door,x,s) ∧ burdened(x) ⊃ should(open(door),s)

	The difference between Doyle's new system and the previous
non-monotonic systems is that his refers to the rule "chivalry" by
name, whereas the others are semantic.  His way of stating the rule
"lib" has the following advantages: If the rule merely stated that
"libber(x) ⊃ ¬should(open(door),s)", it would fight with both "chivalry"
and "courtesy" and a further principle would be required to resolve
the conflict.  At least in this case, the specific mutual reference
of the rules, expresses precisely what we want to say.

	Some further comments:

	1. A set of Doyle rules, or possibly other non-monotonic
rules, might be "compiled" into a set of predicate calculus rules
but would then lose their ease of non-monotonic modification.  Thus
the three above rules would be compiled into

approaches(door,x,s) ∧ (burdened(x) ∨ woman(x) ∧ ¬libber(x))
	⊃ should(open(door),s),

but this would not admit the further non-monotonic modification

tease:	want(tease(x),s) ⊃ inapplicable("lib").

	2. Doyle's way of handling second and higher order
considerations, has the virtue that going higher peters out
as soon as a unique rule with pros and no cons appears.  Many
other approaches to higher order considerations tend to proliferate
as we go up.

	3. In every known case, weights and probabilities are the
result of higher level consideration of the problem, while the
higher level considerations themselves are carried out without the
help of conscious weighting, and unconscious weighting seems
implausible.  This indicates that Doyle and other non-monotonic
theorists are right in subordinating wieghting to "deliberation".

Note of January 17

	The above is too hasty in concluding that the Doyle's
example (at least in my modified version) requires reference
to the names of rules.  Here is another version where I have
used the McDermott-Doyle notation although circumscription
also works.

1. chivalry ∧ womanatdoor(x,s) ⊃ should(hold-door-for(x),s)

2. lib ∧ libber(x) ⊃ ¬chivalry

3. courtesy ∧ burdened(x,s) ⊃ should(hold-door-for(x),s)

4. M chivalry ⊃ chivalry

5. M lib ⊃ lib

6. M courtesy ⊃ courtesy

	There might be some advantage in parametrizing chivalry,
i.e. writing (1) as

chivalry-to(x) ∧ womanatdoor(x,s) ⊃ etc.

and writing (2) as

lib ∧ libber(x) ⊃ ¬chivalry(x)

and writing (4) as

∀x.(M chivalry(x) ⊃ chivalry(x)).