COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Comments on a Creary draft
C00011 ENDMK
C⊗;
Comments on a Creary draft

line 43
example, the imperative advice "Place greater emphasis on the propagation
and use of non-mandatory constraints", given to a planning program, might be
expressed formally as "Intensify-related-activities( --, --).  The forms

This is too vague either to express the adviser's thought or to be taken
as advice by the advice taker.  Perhaps my problem is that the advice is
so global as to be rarely appropriate.  Maybe it would be more nearly
appropriate if its scope were restricted to a particular search or a
particular kind of search.  Maybe this was intended.

line 63

  2.  Modularity of design of component problem solvers is to be the key
      to advisability of these problem solvers.  This modularity will
      result from the structure of the high-level language in which the
      problem solvers are written.  

It seems optimistic to suppose that the modularity will result from the
structure of the high level language.  The most that can be hoped for
is that the high level language will admit the desired modularity.  The
programmer, man or program, will have to provide the modularity.  Alas,
modularity seems to require a special effort aimed at modularity.  Just
thinking about the problem seems to result in non-modular programs.

line 66

  3.  When advice taking requires a restructuring of a component problem
      solver, this is to be accomplished by a (usually only partial)
      reprogramming of the component by a specialized planning program
      that is of the same basic design as all other component problem
      solvers.  The hierarchical nature of the plan (i.e., the program)

This is the famous GPS wish.  It remains to be seen whether this
new formalism admits it.

line 98 - I'm not convinced that deductive methods are inappropriate.
However, since they will probably involve non-monotonic reasoning,
the correctness of their results will not be absolute.

line 154

	We choose not to cast this
program as a theorem prover, out of a feeling that automatic deduction is
but one specific kind of problem-solving task among many, and that more is
likely to be learned about the different kinds of problem solving and what
they do and do not have in common if no blanket working assumptions are made
at the outset about the reducibility of all problem solving to the
specifically deductive kind.  Our alternative is to cast the problem-solving

I still like the deductive formalism and think that a fundamental problem
is to put the other heuristics in a declarative form.

line 177 - The preference for deduction is not just to make it convenient
to produce rationalizations.  That isn't important.  I suppose that the
farther we are from deduction, the less likely it is that the process is
self-applicable.

lines 192-203 - These points seem wishful to me.
The ideas for their realization are not stated, and I worry about whether
they are available.

line 235 - The exposition of Stefik's ideas leaves me wondering how wide
is the domain of their applicability.

page 5

line 10

The qualification and frame problems, which arise when one attempts to
infer the cumulative effects of operations for planning purposes, are to be
handled in a unified manner by the combination of a knowledge-based,
resource-limited inference procedure and two fallible rules of plausible
inference.  When the task arises of inferring the problem-relevant aspects

This characterizes the frame problem, but it doesn't seem to me that it
characterizes the qualification problem in general, i.e. the conclusion
that there isn't a bridge in the missionaries and cannibals problem isn't
a determination of the effects of an action.

line 20

parameters of the situation that will remain unchanged (we shall call these
the NEGATIVE inferences).  The problem-relevant aspects of the situation are

I follow you, but this isn't what "negative inferences" will suggest to
most people - it will be taken as referring to deletion of assertions
about previous situation or the attachment of ¬ to sentences.

line 25

to be specified in a data structure called the "problem frame," which is to
be constructed by the problem solver specifically for a given problem, on

The idea of constructing a problem frame seems good.

line 43 - One need not want to deduce as far ahead in time as possible.
The program might get stuck in a cycle of predicting sunrises.  In general,
this outline of positive inference seem worth the effort of trying to
describe it more precisely.

line 49

is performed, and allocating an amount of computational resources to the
process that is appropriate under the circumstances.  This rule specifies

I think there needs to be a different kind of limitation than by
computational resources.  Subjectively, it seems that computational
resources are rarely the causes of our stopping thinking about the
consequences of an event or action.

line 106

  5.  LISP as a universal formal language for the representation of facts,
      assertions, questions, commands, patterns, goals, plans, procedures,
      programs, etc.  An integrated, systematic semantics needs to be
      provided for all these expressive functions of the LISP notation.

LISP is universal for procedures in the usual sense.  The sense in which
it is universal for assertions is purely syntactic, since no semantics
or logical syntax is provided by LISP per se.