COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	mcderm[f86,jmc]		Notes on McDermott's Critique of Pure Reason
C00012 ENDMK
C⊗;
mcderm[f86,jmc]		Notes on McDermott's Critique of Pure Reason

Remarks:

1. I think Hayes's Naive Physics Manifesto is on the wrong track, but it isn't
obvious how to set up a general logical framework for naive physics.
My criticism is that basing the entire formalism on the space time
tubes unnecessarily encumbers most common sense physics reasoning.
I prefer ``common sense physics'' to ``naive physics'', because
common sense physics doesn't have to be naive.  Actually both
terms miss the point which is the epistemological situation.
Common sense physical reasoning must make do with an epistemological
situation characterized by

	a. No clear boundary between the phenomenon being observed
and the rest of the physical world.  To the extent that such a boundary
can be established it is a result rather than a premiss of the
common sense reasoning.  Non-monotonic reasoning is used here by
humans and will be essential for machines.

	b. Lack of quantitative knowledge of the physical characteristics
of the situation in which reasoning and perhaps action is required.

	c. Lack of precise physical theory relevant to the situation.

	d. Inability to do the numerical computations that would be
relevant even if the information were available.

	Physicists' physics sometimes provides an island in the
sea of common sense physics of a phenomenon.  If we contemplate
dropping a barometer off a building and have a watch, we can use
the fact to tell how tall the building is.  However, the relevant
physics formula $s=1/over 2 gt↑2$ must be selected by common sense
physical reasoning which also controls the time observed from the
watch and the computation of $s$.

	Hayes's attempt is epistemologically inadequate as a framework
for common sense physics.  Developing an adequate formalism will
be a work of years and will probably require a touch of genius.

	The relevance of this criticism of Hayes and general remarks
about common sense physics to McDermott's lament is the following.
As long as logic is the only game in town, and it really is, repeated
attempts will be made to use it.

2. McDermott's main summary of the logicist position is correct as is the
presumption that deduction plays a significant role in getting
conclusions from facts.

	The point about writing down what people know apart from
how it is to be used is tricky.  First, no-one has actually succeeded
in doing that, because no-one yet as a sufficiently general formalism.
All the blocks world axioms presume a very special epistemological
situation.  Second, there are still serious unsolved problems even with the
formalizations of toy worlds.

3. The inference from the empty soda cup to the cause indeed includes
something other than deduction.  However, deduction is also
an essential part of that reasoning.

%We agree with the snide remark that logic programming is often
%mistaken for logic.  However, the situation requires clarification
%and not merely wisecracks.
%
4. The remark on p.4 about not proving plans will work is misguided.
Within a model of the situation, one indeed does something very
like proving a plan will work.  If the reasoning within the model
needs to be elaborate this may constitute most of the thought.
However, devising the correspondence between the model and
a certain aspect of the world and convincing oneself that the
correspondence is adequate for one's purposes indeed involves
other kinds of reasoning as well as deduction.

Now the defenses.

1. The idealization defense.  This defense is correct, except that
the processes whereby an idealization is established and maintained
need to be understood in their deductive and non-deductive aspects.

2. The vocabulary defense.  It is indeed true that it is desirable
to give words technical senses in common sense reasoning formalizations.
Ordinary language words are appropriately used as mnemonics, and it
is ok that the technical terms don't carry all the connotations
of the natural language word, provided no-one tries to sneak the
extra connotations back in later without appropriately modifying
the axioms.

3. The queen of the sciences defense also applies.

4. The ``metatheory'' defense.  This isn't a defense, because
the object level theory has to be ok too.  Metatheories will also
be important, especially when we want systems that discover.

5. The ``deducto-technology'' defense.  The existence and utility
of logic programming is a good clue, but logic programming doesn't
and can't provide all the logic intelligent computer programs
will have to use.

6. The ``non-monotonic'' defense.  Formalized non-monotonic reasoning
is less than ten years old, and there is very active exploration of its
possibilities for extending formalisms and reasoning.  Since McDermott
wrote this opus, our group alone has produced two new formalisms that
solve the ``Fred's death'' problem, i.e. give earlier events suitable
priority.  Lifschitz's latest formalization is as simple as can be
asked and translates directly into logic programming.
If McDermott works at it, he can undoubtedly come up with some
new problems that will give difficulty.  While this one was solved
in a few months in a quite general way, he may be lucky and find
a really difficult problem that will take years to solve.
We logicists will keep at it, however, and the only real way to
beat us is to think of a different way of doing things that is
a lot better and isn't just logic in disguise.  As soon as the
people who attempt to devise a new formalism that is capable
of expressing general common sense knowledge and also expresses
knowledge in a communicable form, I'll bet they'll find themselves
reintroducing most of logic.