COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	comments on Sloman's "Computer Revolution in Philosophy"
C00032 00003
C00033 ENDMK
C⊗;
comments on Sloman's "Computer Revolution in Philosophy"

xi I don't see that the essence of biology concerns computational
processes, although these will be important for understanding the
biology of the nervous system.

chemistry:physics = programming:computer is not the right analogy.
Better is biology:(physics+chemistry) = programming:computer.
Many programs are compatible with a given computer and many
biologies are compatible with (physics+chemistry), while presumably
there is only one chemistry compatible with physics.

a certain lack of respect for the free market - perhaps

xii The arguments about whether machines can think are sensible,
although Sloman is quite right in avoiding the matter if he
has nothing to contribute to it.

xiii I wouldn't characterize Weizenbaum as racist.  We humans have
an entire right to refrain from building robots or building only
subservient robots if we suppose that this is in our best interests.


1 A certain amount of fashionable anti-technology.

2 He likes the term metaphor.  To me speaking of metaphors, suggests
that an idea is being proposed that will not be seriously defended
if attacked.  Metaphors should be discussed only in connection
with the most tentative ideas, and one should replace them by
precise ideas or definite analogies.

Good new point about the cruder mechanical metaphors.

3 Good point about there being more to science than equations.

4 Good list of philosophical-AI questions.

5 It is all very well to praise AI for raising questions, but
the proposals for solving them are almost always extremely naive
and probably no present proposals are sophisticated enough.  Therefore
the work of Dreyfuss and Haugeland, even if done from a negative
point of view, can be helpful.

"Philosophy can make progress, despite appearances.  Perhaps in future
the major advances will be made by people who do not call themselves
philosophers".

6 Methinks he relies too much on Boden, especially as doesn't mention
Boden's critical reservations about much of the work she describes.

7 As to purpose, send him "Ascribing ...

"So biologists and psychologists who aim to banish talk of purposes
from science, thereby show their ignorance of some of the most important
new developments in science".

8 Good crack at cybernetics and systems theory and information theory.
It is important to point out the irrelevance of these theories, and
I think he is right about the nature of their limitations.

9 The information theorists can say that we do decode - using background
knowledge is provided for in the information theory model.  It is their
specific mathematics that is irrelevant.  Fortunately most present
coding theorists now apply their work where it is relevant.

"Thus reductionism is refuted".  If that is what the reductionists meant -
but maybe it is what they meant.

10 Good. It is indeed the job of philosophers to criticize inadequate
concepts used in AI.  However, there are two distinguishable forms
of inadequacy. (1) the work covers only a limited domain.  That's
ok if recognized, and the work may still be valuable even it its
author doesn't recognize its limitations and someone else has to
do it.  (2) The foundations of the work are on quicksand, and it
doesn't even to what it claims on the examples it does.  E.g. Colby's
program is too stupid to really model the mind of a paranoid.

13 "Progress in philosophy (and psychology) will now come from those
who take seriously the attempt to %2design a person%1.  I propose a
new criterion for evaluating philosophical writings: could they
help someone designing a mind, a language, a society or a world".

15 Good remarks about common sense vs. common opinions.

18 I would claim that emotional and intellectual abilities are
separable.

28 I don't think that even quantum mechanics is an exception to the
rule that the laws of physics are not statements of probabilities.
While the laws of quantum mechanics tell us that some things can
only be known probabilistically, the laws themselves are not statements
of probabilities but rather laws of motion of wave functions which
generate probabilities but aren't themselves probabilities.  What
they are is even more mysterious.

33 Sloman has become a very clear thinker.  It is rare to find someone
who doesn't smooth over the things he doesn't understand and can
leave the loose ends sticking out for the reader.  Bravo for that.

35 Glad to see somone who hasn't joined the Kuhn bandwagon.

37 "Usually philosophers plunge into discussions of such questions
as whether we can know anything about the future, or rationally
believe anything about the future, without first asking how a
rational being can even %2think%1 about the future or ⊗think about
alternative possible future states of affairs".

Is Piaget really good for something?  I find him totally muddled.

38 Good set of examples of finding and improving concepts.

40 Good list of ways of criticizing a typology.

49 Fuels are used at efficiencies greater than those of animals.  What
hasn't been duplicated is to do it without high temperatures or
pressures with such light energy conversion systems.

52 Only certain kinds of theories - theories of how systems evolve
in time - are computer-modelable in an a priori defined sense.  Even
for these theories, there are qustions modelling can't answer.  Thus
the three body problem is modellable, but running the models can never
give the conclusion that such a system is stable.  To take another
example, the use of the computer in proving the four color theorem
did not simply take the form of generating maps and coloring them.

50 Good set of desiderata for theories.

58 Defense of partially unrefutable theories.

59 I agree with Sloman's emphasis on possibitity, but I suspect it should
be formulated somewhat differently.


********
Starting again on p. 104

104 Godel's theorem is relevant to the limitations of computers.  These
limitations apply to people also.  The closed system is not the problem.

105 I don't know of substantial examples of programs that work for reasons
their designers don't understand.  Failing for reasons their designers
don't understand is entirely common.

107 I am not sure that the issue of materialism vs. idealism is so readily
dismissed.  Admittedly answering that question doesn't answer all other
questions, but it still remains.  Perhaps I am mistaken, but I suppose
that idealists give spirit some non-causal properties, i.e. they wouldn't
accept a PDP-10 made of spirit as a mind.

109 While Dreyfus is wrong in demanding that computers be made of meat,
the argument that present computer programs play chess differently from
humans is ccrrect.  They examine far more positions but lose much of the
benefit, because they don't explicitly recognize the patterns in positions
or the spatio-temporal patterns of strategy.  When we understand these
things better, we shall write better programs.

112 "In both cases the whole is far more than the sum of its parts" is not
an explanation but mere rhetoric.

116 It is hard to see that calling the mechanism part of the environment
and vice versa improves the description of their interact which, by the
way, is not that of mutually recursive subroutines.  An organism and
the environment give each other information, but neither is compelled
to reply in the manner of return from a subroutine.

120 "This is almost certainly incompatible with assumptions made by
economists and some moral philosophers.  For instance, there need not be
any overall tendency for the rules to optimize some abstraction called
'utility`".

The differences between the human motivational system and what we might
be inclined to build have not been noted, i.e. the effect of chemical
state on logical processes and the escape of subgoals from the main
goals.

....
Sloman is a Francis Bacon, but his lists are more useful than Bacon's were.
In any case, he is not a Newton.
....

127 I was expecting an explanation of how the administrator is not a
homunculus, i.e. an explanation of how it doesn't have to contain
all the intelligence of the system within itself. - see p. 115

"However, all that a %2physical%1 sense organ can do is produce some
kind of spatial or temporal array or manifold of physical values (as
a television camera or microphone does).  This does not yet amount to
perception: it simply amounts to the productin of a new structure within
the system.  Whether anything is thereby perceived, and what is perceived,
depends on what procedures (or programs) are available for analysing the
new structure, finding relationships between its parts, prehaps
manipulating or modifying it (for example, correcting misprints or other
errors), interpreting it, and making use of all this either immediately
or later on in the performance of actins or solving of problems". - good
but it omits "identifying parts" in the above lists.  Probably there
are other omissions too.

128 "So the objective/subjective distinction evaporates". - wrong.  It
still makes sense to ask what an organism has discovered about the
world external to it.  Remarks on meta-epistemology.

129 (retina) "sends interrupt signal"  This is unlikely to be literally
true of biological organisms, so we need to ask what really happens.

133 the conclusion that even new-born infants must have a rudimentary
concept of self seems dubious. (1) a haggle - unless the new-born shows
abilities that require it, the concept of self may result from maturation
before its use is demonstrated. (2) more substantial - The simplest form
of learning is the adjustment of parameters.  It is hard to see that this
requires a concept of self.  Secondly, the concept of self undoubtedly
fragments, e.g. "I am a human like the others" is simpler than analyzing
one's consciousness, and more complex than "a new procedure for crossing
streets is needed".

136 "The normal approach in social science is to shirk the task of
understanding such dependencies, by representing the transitions as
probabilistic ..."

141 "Clearly, the task of interpreting and diagnosing pathological
behaviour in such a complex system must be extremely difficult.  It
cannot be done without a good theory of the normal structure and functions
of the system.  This is why I have little faith in current methods of
psychotherapy".

various caveats about the incompleteness of his description

143 he denounces flow charts, but obviously he has no better way of
describing things to laymen.  Perhaps there is a way of describing recursive
procedures for laymen.

144 It is certainly possible to devise logical systems in which
diagrams play an essential role.  It is necessary to have a "Euclidean eye"
which can determine that a diagram meets certain specifications and
can observe relations in the diagram.  It can be proved about Euclidean
geometry that certain relations true in a diagram must be true in all
diagrams.  Presumably the algebraic geometric concept of generic point
can be used.  However, the most obvious form of Euclidean eye requires
infinite precision of measurement.  With further mathematics, circumstances
can be given under which any relation (of limited complexity) true to
five decimal places is exact.  No-one has given such a theory.  The
topology of plane graphs is an area in which the use of diagrams to
make theorems is less problematical.

	It would be nice to devise some analogical representations that
clearly beat logical deduction or else prove that anything doable by
analogical representation in a practical way can be done by deduction
with linearly proportional work.
Some precise mathematical problems can probably be formulated here.
The theory of NP-completeness tells us that whole classes of problems
has polynomial relationships in their sizes of proofs.

The ability to show that an object can be moved from here to there by
moving a model in a diagram or even in a three dimensional model of
the environment is difficult enough to program so that it may be
a candidate for useful analogical representation no easily digitized.

145 I hope McCarthy and Hayes didn't say that logical inference is
the only inferential mechanism that can be useful in AI.

157 "Our main lack at present is not data so much as ideas on how to build
suitable theories".

159 "So validity is a semantic notion, concerning meaning, reference,
and truth or falsity, not a syntactic notion, as is sometimes supposed
by logicians.  They are led to this mistake by the fact that it is
possible to devise syntactic tests for validity of some inferences,
and indeed the search for good syntactic criteria for validity has been
going on at least since the time of Aristotle". - Mathematical logicians
have understood this point since the late 1920s.  It would be hard to
find a textbook in mathematical logic in use today that doesn't fully
recognize it.  Sloman would benefit from studying mathematical logic.
[Maybe he studied logic from Nidditch].

161 It is all very well to say that specifying the semantics of `red' requires
referring to the psychology and physiology of color vision.  In some
sense this is true, but then it turns out that no person today can
specify the semantics of `red', because no-one knows enough about
psychology and physiology.

Well I suppose my objection is met by his subsequent qualifications.

165 Do all analogical representations have parts which denote parts
of what they represent?  Maybe.

 ......
Consider a graph and ask if we can represent it as a linear graph in
Euclidean space in such a way that it is possible to go from any
vertex to any other by going towards it in the Euclidean space.
 ......

 ......
I like his emphasis on studying what people can do rather than trying
to predict what they will do.
 ......

191 "I do not believe that educational psychologists have even the
foggiest notion of what such a mechanism might be like ..."

200 dubious notion of creativity

"When computers are programmed to know so much they will be just as
fallible, and they will have to improve themselves by the same painful
and playful processes we use".  When computers are programmed to
use circumscription, they will make errors of jumping to conclusions.
However, there is no reason for these errors to occur in those
computations not requiring circumscription, so they won't make
errors in domains in which they don't today.

211 Belief can be analyzed without analyzing desire or decision.

213 "Piagetian psychologists comment on some of the achievements, but
provide no means of analysing or explaining the underlying mental
processes discussed here".

215 "The old nature-nurture (heredity-environment) controversy
is transformed by this sort of enquiry".  Again false.  Sloman hasn't
disposed of as many issues as he thinks.  What does his transformation
say about the justification for reverse discrimination.

221 "A theory which claims that perceptual achievements are not
decomposable into sub-processes cannot be used as a basis for designing
a working mind which can perceive any of the things we perceive".

Sloman's reluctance to quibble about whether the word "knowledge" should
be used for talking about the stored facts and procedures used in
perception is failing to make the distinction between knowledge and
ability, and is just the kind of thing he inveighs against in
the previous paragraph.  Its only justification is that it helps
his side in the controversy over whether Kant was right - a matter
not of current interest.  Well he softens it.

226 "Our present ignorance is not a matter of our not knowing which theory
is correct, but of our not even knowing how to formulate theories
sufficiently rich in explanatory power to be worth testing experimentally".

240 good about the need to list things that computers can't do yet

242 "how is it possible for there to be a distinction between conscious
and unconscious mental processes?"

244 this box does this and that box does that, and the seventh box is
the one that tells all the others what to do

247 It requires special provisions for internal information about
the state or actions of the system to become data for the system.
Therefore the problem of consciousness is how are some internal
observable - not why aren't they all observable.

250 he obviates previous comment

252 bottom Also the ability to describe a scene may be one that
can be acquired and likewise the ability to interpret such a
description.  It would be interesting for two people to practice
communicating scenes over the telephone - one looking and describing
and the other drawing - allowing also questions.

253 bottom Sloman is much too hopeless about the prospects for
developing techniques for observing the brain or computer programs.
The need and the possibilities are particularly acute for game-playing
programs which are discussed as though they were unobservable human
players.

 .....
It would be worthwhile to consider other cases of scientific information
that doesn't take form of predictions..
 .....

259 Dubious notion of aesthetic experience as an epiphenomenon of
complexity.  Sloman also seems to regard emotion as an epiphenomenon
of complexity - equally dubiously.

260 Dubious ascription of creativity to present programs

262 The nth Marxist slogan, where n is excessive.

I don't see in what sense the word "rational" is being used.

267 The ability for subgoals to escape their main goals is a
characteristic of humans that might be wanted before machines
could be called free.  I am against making machines with this
characteristic.

I'm not sure about the identification of depth-first with optimistic.

272 It all ends in a burst of anti-Earthman propaganda.
Sloman's anti-Earthman prejudice leads him to skip steps in coming
to the conclusion that intelligence requires human-like motivational
structures.

285 Sloman's postscript on the non-need for hierarchy of meta-languages
shows a lack of knowledge of logic; he doesn't understand what mathematical
theorem Tarski proved and what further philosophical claims it has led to.
Sloman's POP-2 program will not begin to decide arithmetical sentences
or even sentences about POP-2 programs; it will run out of storage on
easy cases.