From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!europa.asd.contel.com!darwin.sura.net!Sirius.dfn.de!zrz.tu-berlin.de!cs.tu-berlin.de!news.netmbx.de!unido!mcsun!news.funet.fi!sunic!seunet!ku Thu Feb 20 15:21:42 EST 1992
Article 3826 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!europa.asd.contel.com!darwin.sura.net!Sirius.dfn.de!zrz.tu-berlin.de!cs.tu-berlin.de!news.netmbx.de!unido!mcsun!news.funet.fi!sunic!seunet!ku
llmar!pkmab!ske
>From: ske@pkmab.se (Kristoffer Eriksson)
Newsgroups: comp.ai.philosophy
Subject: Re: The Putnamn FSA (was: Strong AI and panpsychism)
Message-ID: <6580@pkmab.se>
Date: 15 Feb 92 10:13:37 GMT
References: <1992Feb6.113740.2533@arizona.edu> <1992Feb8.033821.16351@news.media.mit.edu> <1992Feb10.000321.26668@organpipe.uug.arizona.edu>
Distribution: world,local
Organization: Peridot Konsult i Mellansverige AB, Oerebro, Sweden
Lines: 125

In article <1992Feb10.000321.26668@organpipe.uug.arizona.edu> bill@NSMA.AriZonA.EdU (Bill Skaggs) writes:
>First of all, I have been informed (by Mikhail Zeleny) that
>Putnam's argument (which can be found in his book "Representation
>and Reality") refers to FSA's rather than Turing machines, and
>exploits the fact that no real object is ever in the same
>state twice to remove the need for time-dependence of the 
>mapping.

As I understand your explanation, this is a very simple mapping, wherein
each successiv state of the abstract FSA is mapped to a new, unique
sequential state of the "real object", never using the same state again
even when they refer to the same state of the abstract FSA.

A have lots of critique for such a technique.

Firstly, how can anyone guarantee that a real object will never be in the
same state twice? What is that claim based on?

Why should the object not enter the same state again if it is allowed to run
long enough, possibly a very very long time, but nevertheless the same state?
And if the states do repeat, then what happens with this simplistic mapping,
if it does not involve a time-dependent component?

Furthermore, even if one state repeats itself, you would probably not see the
same sequence of states follow after that state again, unless you add
precautions to isolate this real object from unwanted outside interference
and eliminate all randomizing effects from the quantum level of reality. And
if this repeated state does not proceed to the same next state as the previous
time (under the same conditions), then it will of course this time violate the
state transition specified by the abstract FSA.

At the same time, I would guess that these interferences and randomizing
effects, are exactly what is being counted upon for to make the object
pseudo-never repeat itself (to no avail). If they are eliminated, and the
system is properly isolated from its surroudings, as normal FSA's should be,
in order to make sure that the system behaves deterministically when it
eventually does repeat itself, then we will at the same time have made the
available states much fewer, and the simplistic mapping falls apart even
more, since it will now repeat much sooner.

There are some other strange properties with this kind of mapping too. If
the real object in question is (again) dependent upon quantum randomness
and external disturbances to keep the states from repeating (whether that
works or not), then it is not possible to spell out the mapping before-hand.
All these mappings are ad hoc constructions after the fact (after the run).
Quite different from "ordinary" FSA mappings. And it is not just that you
may not yet have had the opportunity to determine how this object works
(which could happen with an "ordinary" FSA), rather, it is *impossible*
to tell the mapping in advance.

Actually, there *is* no mapping in advance, because the mapping depends on
which moment the FSA is said to be started, since the starting state will
be different. I don't find this very time independant. And also, you can't
predict at all which internal states the object is going to go through,
whether viewed as an FSA or otherwise.

And not only that, even after having run it once (or any number of times),
you can still not tell what mapping it will use at the next run, unless
there suddenly really does exist a way to make it repeat earlier states
(on command), violating its basic assumption.

You can't even describe the total mapping at all. The mapping is infinitely
long, mapping an infinite number of states of the real object, to the
abstract FSA, if you take all possible future runs into account. And the
mapping does not repeat itself, leaving no possibility of higher-order
finite descriptions. Thus I don't see any possibility of actually spelling
out the total mapping, at least not in terms relevant to the object itself
(rather than for instane "the 1st state of the 19th run" and such), in
bright contrast to ordinary FSAs. In addition to the fact that there will
never be a time where the total mapping is determined.

Even if the FSA does not run for very long, and in stead just enters the
"halt" state soon, the problem of eventual repetition is still there, since
the total mapping has to cover all subsequent runs of the FSA too, if one
wants to claim that the object itself implements this FSA (and the object
is the same in every run), rather than acknowledging that we're just placing
a new arbitrary one-time mapping onto the object anew for each separate run,
putting the mapping in focus rather than the object.

Thus, based on the description given, I think the whole Putnam idea rests
on a false assumption, and additionally has very many properties that are
radically different from ordinary FSAs, making it very doubtful that it
is a valid FSA mapping at all. I expect that even more definitive arguments
could be given, if it was worked through some more.

I think for instance, that not having the object properly isolated from
undue external influences, is quite contrary to the idea of an FSA, and
as far as I can see, this whole method is dependant on such undue influence
to keep it from repeating states, but I don't have good enough a definition
of FSAs to rule it out immediately.

What I mean by undue, here, is that these influences are not inputs, yet
they still influence the state transitions of the object in such a way that
if they had been changed slightly or removed, then the object would have
passed through completely other sequence of states, that would no longer
have mapped to the abstract FSA through the same mapping. Actually, these
influences determine the mapping, rather than the reverse: that the mapping
plays down the consequences of these influences in such a way that they
don't make any difference to the run.

Thus I believe any "real" FSA should ignore all influences except its
specified inputs (and possible device failure outside its operating
environment limits).

I also think it should be possible to specify the total mapping in some
finite way, although the total number of possible states being mapped
might be considered infinite when looked upon with infinite resolution.
This also implies that I think the mapping should be the same from run
to run (or reversely: the same mapping should cover all runs).

I also think that it should be possible to reverse-engineer the state
transition table of the abstract FSA from a description of the mapped
states (but not the transitions) as they appear on the object, together
with knowledge of how the object behaves in its own right based on lower-
level principles, like physics for instance. This is impossible with the
Putnam device, as I understand it, since it is not possible to predict
accurately what it is going to do before-hand at all, due to external
influences, and the mapping is not available before-hand either. It would
be possible to reverse-engineer a particular run, after the run, but not
based on lower-level principles alone, without any run at all.

-- 
Kristoffer Eriksson, Peridot Konsult AB, Hagagatan 6, S-703 40 Oerebro, Sweden
Phone: +46 19-13 03 60  !  e-mail: ske@pkmab.se
Fax:   +46 19-11 51 03  !  or ...!{uunet,mcsun}!mail.swip.net!kullmar!pkmab!ske


