.require "let.pub[let,jmc]" source
∂AIL Professor Hilary Putnam↓Department of Philosophy↓Harvard University
↓Cambridge, Massachusetts 02138∞

Dear Hilary:

	I found your lecture quite stimulating, although I have to confess
that mainly I was stimulated to disagree.  I hope you won't think I always
disagree; indeed I found much to agree with in some of your earlier papers
to which I referred in my %2Ascribing Mental Qualities to Machines%1.

	First, and this may reflect only my philosophical inexperience, I
wasn't convinced that there was a real problem about reference.  There is
a relation between the outline drawn by the ant and President Carter, and
it is legitimate to study the relation between pictures and the objects
they look like independent of anyone's intention.  Such a theory would
tell us when a cloud or ink blot or rock formation or pattern of chicken
entrails would be found to be a picture.

	Likewise, there is another theory of systems of representation
that would cover both linguistic and pictorial representations.  Such
systems might be used by either people or machines.  Many questions about
whether an alleged symbol represents something may be better asked in
connection with a specific system of representation rather than as
intrinsic questions.

	It seems to me that the possibility of asking what a symbol
represents in isolation from a particular system of reference is that it
almost never turns out that a large sequence of symbols admits more than
one interpretation.  This is suggested by experience with cryptography.
So far as I know, there are no known examples of cryptograms that admit
more than one lengthy interpretation in ciphers with key length
substantially shorter than the message.  Although you didn't spell out
your example of something that could be interpreted as representing either
prime numbers or apple trees, I doubt that there is such an example in any
non-trivial sense.  If this is right, however, there won't be any good
simple examples, because the uniqueness of reference will depend on having
a large amount of data.  Therefore, it seems to me that the theory of
reference systems will be more fundamental and much more tractable.

	Maybe there are even more ways of looking at representation.
It seems unlikely that there will be a single true concept of representation.
Moreover, it seems unlikely that examination of ordinary language will
dictate preferring one of them, especially if it turns out that the
differentiating examples are outside of ordinary experience.

	The problem of the %2vat men%1 brings up matters to which I
have given considerable thought though mainly of a science fiction
character.  Might we be simulations in some giant computer, and could
we investigate the question?

	It seems to me that the perfect simulation is, as you mentioned, a
limiting case of imperfect simulation.  Moreover, while investigating the
matter, the alleged simulee might say that the simulation, if such, seems
perfect so far, but he still hopes for a flaw.  One flaw I imagined was
that a digital simulation of differential equations might show round-off
error, and that such errors would seem random when looked at
superficially, but would betray their digital nature when examined more
closely, i.e. the different errors would show correlations and other forms
of non-independence.  My fantasy went on to suppose that the program was
huge, i.e. corresponded to millions of years of human work, and embodied
billions of arbitrary decisions among different ways of accomplishing the
same purpose, each of which gave a clue to the psychology of the
Programmer.  From this, again using cryptography, we might conjecture
properties of the "real world" and how we might act in the "real world".

	While listening to your lecture, another way of detecting
simulation occurred to me.  It is problematical whether it could be called
looking for an imperfection.  To put it in old-fashioned terms, we could
look for %2final causes%1, i.e. we could attempt to make a theory that
would explain apparently random connections among phenomena as
consequences of the purposes of the Programmer.  Since the time of
Galileo, final causes have quite properly gone out of fashion.
However, they still have a small domain of legitimacy.

	When we find that an individual or group in history performed some
action or had some custom, it is often helpful to ask why they did it that
way.  When we see an organ on an animal, such as the pit of a pit viper or
the comb of a cock, it is often helpful to ask what function it might
perform.  We may or may not get a useful answer.  On the other hand, it
seems unfruitful to ask the purpose of a range of mountains along the west
coast of the American continent.  It seems to me that the inutility of
asking for final causes outside the limited domain of human purpose and
natural selection is an empirical fact; it would not have been legitimate
to make it an %2a priori%1 philosophical principle.

	Indeed, it is common for people to ascribe the course of
political events to purposeful behind-the-scenes action whether it
be the unfolding of God's grand plan or the evil machinations of
the Vatican or British Intelligence and the Rockefellers (the latter
being the current doctrine of the U.S. Labor Party and its maximal
leader).

	From an aesthetic point of view, it may be unfortunate that it is
unfruitful to explain the ant as an admonishment to be diligent or
the rainbow as a sign that God will destroy the the world with fire
next time, i.e. that
.begin nofill

%2God gave Noah that rainbow sign.
God gave Noah that rainbow sign
That it won't be water
But fire next time%1.
.end

does not fit into a larger theory that allows successful prediction
and control of phenomena other than the behavior of people who
believe the theory.

	Indeed there is another science fiction fantasy that suggests
a form of astrology.  Suppose that there is enough matter in the
universe so that it will contract again, and everything will be
burned up in the compression which will be followed by another
big bang.  If the compression is merely to a high density and not
to an informationless singularity, then large scale inhomogeneities
in the present universe will lead to inhomogeneities in the
large scale distribution of matter after the next big bang.  If we
want to "survive" the compression, we might, having nothing better
to do for a few tens of billions of years, encode our own descriptions
in the large scale distribution of matter in the universe, i.e. by
rearranging the galaxies.

	When life evolves after the next big bang, and scientists study
the large scale distribution of matter, they will notice that information
is encoded in this distribution.  If we make a plot analogous to that of
Fred Hoyle's %2A for Andromeda%1, this information would contain
instructions on how to build a computer.  If they foolishly built this
computer without precautions, it would take over their society and use its
resources to read the rest of the message and reconstruct us.  Of course,
we can ask whether there is a message in the distribution of galaxies or
in the digits of the dimensonless constants of physics left over from
before the big bang.  I suppose that in the next few billion years, even
such improbable possibilities as this will be checked out.

	The third, and admittedly weak, reason for considering the
vat hypothesis might be philosophical prejudice, i.e. a person might
dislike the big bang hypothesis enough to hypothesize that the apparently
big bang universe is a simulation within a universe that has gone on
indefinitely.  The idea that the real universe is a Hilbert space or
other infinite dimensional space has many attractions for the wishful
thinker.

	Incidentally, when you referred to me in connection with
"procedural semantics" which I took to correspond to empiricism, I
wondered whether you thought that efforts to program machines to behave
intelligently were somehow committed to giving them an empiricist
way of looking at the world, i.e. having them attempt to "construct" their
world as
the "input-output" relationship of the machine's environment.  It
seems to me that this is not the only way or even the best way to
program machines.  My thoughts have always run towards trying to
specify a "naive realist" point of view well enough to program it.
The intelligent program should be ready to accept names as identifying
unknown objects and hope later to learn more about them.  Ideally,
it should have the ability of a child to listen to a poem beginning
%2"The lion and the unicorn were fighting for the crown"%1 and only
later come to realize that a lion is a real beast and a unicorn
a mythical beast.

	I have done some thinking about what I used to call
metaphilosophy and which may be more precisely called metaepistemology.
Specify a world as a causal system, e.g. a system of interacting
finite automata, and suppose that it contains a computer that
can execute programs, and consider programs that "attempt to discover
the structure of their world".  We consider building into the program
various epistemological methodologies, e.g. empiricism.  It is then
a technical question whether the program will indeed discover the
structure of its world.  If metaepistemology could be developed
as a technical subject, it might have the same effect on epistemological
controversies as metamathematics had on the controversies about the
foundations of mathematics.  It didn't actually destroy the controversies,
but it limited them.  It is hard to be a fanatical intuitionist on
discovering that intuitionist and classical number theory are equi-consistent.

	Enclosed are some papers, some of which I may have sent you
before.  They are not mainly a computer scientist's desire to poach
on philosophical territory, even if they may be taken that way.
Rather they are an attempt to tie down some naive philosophical
ideas to the point where there is a chance of building them into
an intelligent computer program.  I am convinced that such a program
has to be provided with some kind of philosophical system, however
naive.
.next page
	I hope you find the above amusing, and I will surely call you
the next time I come to Boston.

.reg