COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%clarke[f89,jmc]		On the Clarke-Collins correspondence
C00021 00003	\section{References}
C00022 00004	\smallskip\centerline{Copyright \copyright\ 1989\ by John McCarthy}
C00023 ENDMK
C⊗;
%clarke[f89,jmc]		On the Clarke-Collins correspondence
\input memo.tex[let,jmc]
\title{Artificial Intelligence and the Clarke-Collins Correspondence}


	The object of this note is to comment on what
contemporary research in artificial intelligence (abbreviated AI)
might have to say relevant to the issues raised in the
Clarke-Collins correspondence.  The answer is not much---but
something.

	One fundamental disagreement between Clarke and Collins
is whether a system can have properties not possessed by its
constituents.  Artificial intelligence, like almost all
contemporary science and engineering, is based on a methodology
that is entirely on the side of Collins.  We are always
contemplating, studying and building systems that have properties
not enjoyed by their parts.

	Moreover, Collins's example of the clock cannot be
improved upon today in arguing for the general principle that
such systems exist.

	As for the mind and consciousness, there is a lot
known that wasn't in the eighteenth century.

	Clarke's argues (p. 730)

{\narrower
In reality, That the  Soul cannot possibly be {\it Material}, is
evident not only from the consideration of its noble Faculties,
Capacities and Improvements, its large Comprehension and Memory;
its Judgment, Power of Reasoning, and Moral Faculties; which
Arguments have been urged with unanswerable Strength by the wisest
and most considerate Men in all Ages from the times of {\it Socrates}
and {\it Plato} to this very Day; but the same thing is moreover
demonstrable from the single ccnsideration, even of bare Sense
or Consciousness itself.\hfil\break
For Matter being a divisible Substance, consisting always of
separable, nay of actually separate and distinct parts, 'tis
plain, that unless it were essentially Conscious, in which case,
every particle of Matter must consist of separate and distinct
Consciousnesses, no System of it in any possible Composition or
Division, can be any individual Conscious Being: For suppose
three or three hundred Particles of Matter, at a Mile or any
given distance one from another; is it possible that all those
separate parts should in that State be one individual Conscious
Being?\smallskip}

	The second sentence is later generalized, and Collins
correctly takes issue with it.  The first sentence indicates that
Clarke includes in the soul all of what we today consider the
mental qualities of the brain.  He doesn't take the escape
route sometimes taken of regarding most of the function of
the brain as performed by a material system with the soul as
something extra.  Granted that, modern science has
much new to say.

	Neurophysiology has identified some of the structures
that perform some of the functions of the mind, especially in the
sensory and motor areas.  Also relevant is what has been
discovered about the interaction of the brain and the chemistry
of the blood and cerebrospinal fluid in affecting mood.  These
new facts would probably impress an eighteenth century mind.
Nevertheless, Clarke could still take refuge in our lack of
understanding of the neural mechanism of higher mental processes
and postulate a unitary and immaterial soul necessary to carry
them out.  He wouldn't be alone, even today.

	One of the parts of the brain best understood today
is the visual system.  It has been demonstrated how the image
on the retina is mapped first onto the primary visual cortex
and from there onto the secondary visual cortex.  However, on
the secondary visual cortex it is still essentially an image.
No path all the way from the eye to the hand has been worked out.
Therefore, I suppose the homunculus notion, that behind the primary
and secondary visual cortex, there sits the equivalent of a
little man looking at the secondary cortex is not finally
excluded by present scientific evidence.  One could imagine this
homunculus as unitary, i.e. not made of parts, either physically
or in terms of interacting abstract mechanisms.  However, no
contemporary research is based on looking for it.

	However, in so far as mechanisms are observed that merely
transform the retinal image into some other kind of image, they
are regarded today as important but not as candidates for the
whole story.  To be regarded today as a candidate explanation of
vision, for example, a proposal must say how the mechanism is
constructed of more elementary parts.

	Artificial intelligence research and computer science
generally provide plenty of examples of mechanisms that carry
out functions that in the 18th century could only be carried
out by humans and animals.  For example, there are robot arms
controlled by programs using television cameras as inputs.
Some of these programs represent facts about the world as
sentences in mathematical logical languages.  This includes
both general facts provided as axioms and particular facts
computed from the television inputs.  They are candidates
for representing what Clarke might have meant by ``Memory''
and by ``Judgment''.

	Nevertheless, there is a present day controversy
that somewhat resembles the controversy between Clarke and Collins.
Almost all AI researchers study mental processes as
logically complex and propose systems that have some
mental qualities and not others.  On the other hand,
some philosophers regard having a mind as a unitary
property and regard the systems that have some properties
and not others as fakes or at best weak analogs of
real mental properties.  Maybe John Searle's (1984)
position can be described in this way.

	The issue takes a sharp form when it comes to
the question of whether it is permissible to ascribe
some beliefs to a simple thermostat, specifically to
regard the thermostat as being capable of the belief
that the room is too warm and the belief that the
room is too cold.  The thermostat is chosen precisely
because it is a system whose behavior can be understood
physically without ascription of mental qualities.
(McCarthy 1979) argues that disallowing ascription
of such mental qualities to simple systems is like
excluding $0$ and $1$ from the number system on the
grounds that numbers, and indeed the concept of set,
aren't necessary for discussing sets with no elements
or one element.  Putting the analogy positively,
allowing ascribing some beliefs to simple systems
when it accounts for their behavior makes the
whole system of belief ascription more uniform
and comprehensible.

	The opposite point of view is that mental
phenomena are holistic and must be studied in that
way.

	Yet another aspect of the Clarke-Collins controversy
is relevant to AI.  Clarke's wrote that consciousness
must be a primary quality and therefore possessed by all
parts of a conscious being, just as extent is possessed
by all parts of an extended object.  This contention depends on
the doctrine of primary, secondary and tertiary qualities.  The
doctrine is not regarded as coherent today, but it was widely
accepted in that era.

	AI cannot afford to ignore a formalism merely because it
becomes incoherent when one attempts to extend it beyond a
limited domain in which it works.  Everything we know how to make
machines do today has known limitations.  Moreover, research in
philosophy in a large part consists of developing conceptual
frameworks and then discovering their limitations.  It seems
quite unlikely that anyone is likely to come up with an
intellectual framework without limitations, i.e. that philosophy,
especially epistemology, will reach a final form soon.  One can
doubt whether it will ever reach a final form.  Limited
formalisms should be available for use within the limited domains
to which they apply unless there is available a more general
formalism that works at least as well computationally within the
limited domain.

	Therefore, it is reasonable to ask whether a formalism
based on ascribing to objects primary, secondary and tertiary
properties can be useful in certain domains in spite of its
limitations.  There are properties that Locke or Clarke would
regard as primary, e.g. having extent and mass that are inherited
by he parts of an object.  Also there are properties, e.g.
beauty, that today are regarded as in the eye of the beholder,
although many of the eighteenth century secondary properties such
as color are now known to have objective correlates, e.g.
reflectance.  However, they don't have the prominence in
scientific theory that many 17th and 18th century thinkers
ascribed to them.  My tentative opinion is that the formalization
of common sense knowledge and reasoning won't make much of them
either, because general meta-properties, e.g. primary and
secondary, aren't explicitly used in much common sense reasoning
either.

	It is also interesting to ask whether modern science and
artificial intelligence has anything to say about the original
issue between Clarke and Dodwell.  It is hard to be sure that one
has entered the framework of an eighteenth century theological
argument, particularly if one is a non-specialist.  Nevertheless,
since no-one will burn me or even demand that I be fired no
matter what I say about it, I'll venture an opinion.

	Dodwell argued that the soul is intrinsically mortal, but
God keeps it going after death in order to reward or punish it.
Clarke argued that the soul must be immortal, because it was
immaterial and because it didn't fit his concept of God's
goodness to immortalize a soul just in order to punish it.

	I can't think of anything modern science has new to say
relevant to God's goodness, except that theories of the
development of the universe (actually since Clarke's friend
Newton and explicitly since Laplace) don't use any hypotheses
about God's purposes and therefore undermine the whole idea.
However, if we regard the soul as something like a computer
program, neither Dodwell's idea nor Clarke's can be logically
excluded, although neither is supported by current scientific
evidence or by the directions of current scientific research.
What's new and relevant is the modern mathematical and scientific
notion of possible abstract structures and and specifically the
notion of computational structure studied in computer science.

	Dodwell first.  Interactive dualism doesn't give
philosophers much satisfaction mainly because postulating it
doesn't give them a mechanism for free will that would have the
properties they want.  Nevertheless, there is nothing logically
excluded about the existence of immaterial processes that
sometimes interact with material processes.  God could move
the program and data constituting the soul of a person to
new hardware, either material or immaterial any time He
wanted to.

	Clarke's idea of the immortality of the soul is also
logically possible.  God could have created souls in any desired
number.  These souls could contain the part of the program that
will behave either morally or immorally and will or won't accept
Clarke's religion when offered it.  We can imagine that before
the person is born, they constitute a kind of data set that is
inactive.  It might enter the fertilized egg in the form of extra DNA
or as a modification of existing DNA at time of conception.
It would then affect the way the brain develops.
An alternative pseudo-theology is also possible in which it enters
the brain as a collection of synaptic strengths at the time
of birth.

	I'm an atheist and don't think any of this is supported
by any of the scientific evidence needed to give credence to a
theory.  Like Collins, however, I try to distinguish between what
I regard as supported by evidence and what I regard as logically
possible.

	It's hard to say whether any of Dodwell, Clarke or
Collins would be other than horrified by the above considerations.

	I wish to thank William Uzgalis for suggesting this
project and for many illuminating discussions.
\section{References}

\noindent {\bf McCarthy, John (1979)}:
``Ascribing Mental Qualities to Machines'' in {\it Philosophical Perspectives 
in Artificial Intelligence}, Ringle, Martin (ed.), Harvester Press.

\noindent {\bf Searle, John R (1984)}: {\it Minds, Brains, and Science},
Harvard  University Press.
\smallskip\centerline{Copyright \copyright\ 1989\ by John McCarthy}
\smallskip\noindent{This draft of CLARKE[F89,JMC]\ TEXed on \jmcdate\ at \theTime}
%File originated on 28-Oct-89
\vfill\eject\end