From: Lee Corbin (lcorbin@rawbw.com)
Date: Tue Mar 11 2008 - 22:17:35 MDT
Stathis writes
> Lee wrote:
>
>> Well, just between you and me, I stopped being a true absolute
>> functionalist a long time ago. There is the terrible problem of the
>> GLUT which <choke> <damn me> is a <gasp> zombie. It could
>> sit down to tea with you and pass every test. So I seize upon
>> "information flow", and time chauvinism as a sailor thrown overboard
>> seizes upon some floating wreckage.
>
> Perhaps the GLUT would not be a zombie. Besides, a real live GLUT
> capable of passing a Turing test would fulfil your requirements for
> time-dependent information flow. On what basis would you say it isn't
> conscious?
I do not believe that a lookup table can implement consciousness,
because no true, reductionistic, locally-variable calculations
performing information flow ever occur in a device using it.
For ease of description together with good mental imagery,
let's revert to that greatest of all instruments for discussing so
many things: the very, very large Life Board. With you and
perhaps many readers, it will be taken for granted that the
Rules of Conway's Life support consciousness, and even
civilizations on a large enough Life Board.
Each subsequent generation (i.e. state) is computed from the last
generation, and is done so both *locally* and *reductionistically*.
The term means "a small change locally has no global consequences,
at least not right away". The latter term means that any phenomenon
on the Life Board is "merely" the consequence of vastly many
tinier calculations, just as in determinism in physics we suppose that
a tree is "merely" the summed behavior of myriad atoms and molecules.
Now then. Suppose that five minutes of someone's life is implemented
on the very, very large Life Board. Suppose that this five minutes of
experience requires 10^50 generations, just to pick a number out of
the air. Each generation, or state, is completely dependent on just
the prior state, or, as you would write, St -> Su (I like your brief
alphabetical indication of subsequent states following causally from
earlier ones) where (t, u) is simply any pair of adjacent letters in our
alphabet.
However, a GLUT (Giant LookUp Table) is neither locally computed
nor reductionistic. Here is how the GLUT could work. Let Sa be the
first state (generation) of the person's five minute experience. Then,
instead of Sb obtaining from streams of gliders and glider/spaceship
logic and so forth, Sb is *merely* looked up in some vast database
as follows. The entire Life Board, from the top row (1,1), (1,2), (1,3),
(1,4), .... consists of pixels either on (1) or off (0). So the first four
pixels could be 0110, 1010, or any other four bits. For the entire
Life Board, then a generation is an immense bit string, e.g.,
011011001010000011...
Next, we perform a seemingly irrelevant step: the string is hashed
by a one-to-one algorithm into some other string of bits. See
http://en.wikipedia.org/wiki/Perfect_hash_function .
This step destroys the "locality", i.e., as above, one small pixel
change on the board will now cause a totally different output
from the perfect hash function.
This new string is then used as an address to fetch the next state from
the GLUT. *Whatever* is at the hash of 011011001010000011...,
used as an address, is loaded onto the Life Board. But you see, there
really is no calculation! No small streams of gliders send signals from
one "organ" to another, no local group of pixels causally evolves.
The "next state" is now thus perfectly arbitrary in a sense.
You might as well have a succession of frozen states, each actually
having no intrinsic connection with the rest (for an intuitive description
of "The Problem of the Succession of Frozen States", see
http://www.leecorbin.com/SFS.html).
Of course this reduces to the problem of "The Theory of Dust", and
we are right back to wondering how a pattern found in one cubic
lightyear of dust, that appears to be Sa, could really be connected
in any meaningful way with another pattern found in another cubic
lightyear 10,000,000 parsecs away. Again, I just don't think that
all those patches of dust constitute consciousness (no information
flow, no time involved). A perfectly consistent position for a
time chauvinist like me.
Thus, so far we have a GLUT that perfectly simulates (but not
emulates) a given person for a five minute stretch of his life. It
*simulates* his life because it only appears to be a conscious
run. As I've in essence argued above, this run is no more conscious
than is a succession of frozen heads, or a movie showing a speech
by Humphrey Bogart that used, say, a million frames per second.
No matter how alive and how conscious Bogart appears, he's
not being emulated at all, merely simulated.
Perhaps in another post, I will have to say how the above can be
extended in a TRULY HUGE GLUT to encompass an intelligent
reply to every possible question that you may put to it. In effect,
it merely "looks up" the answer. Unlike Searle's Chinese Room,
however, it doesn't manipulate information or perform calculations.
The lookup is merely indexical.
Thus I say that although it easily passes the Turing Test --- and
every other imaginable test --- this Truly Huge GLUT is a zombie,
and could in no possible way be fabricated, even theoretically,
with what is available today. It's size would be prohibitive. It would
require some kind of super-computronium. For those who aren't
familiar what that is, see http://en.wikipedia.org/wiki/Computronium
So zombies are to me, only the faintest of ridiculously improbable
developments, and so zombies remain completely impossible except
in super-extreme thought experiements such as this.
Lee
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:02 MDT