Re: Emulation vs. Simulation

From: James Rogers (
Date: Wed Mar 28 2001 - 13:18:00 MST

At 07:34 PM 3/27/2001 -0800, Lee Corbin wrote:

>But so far as I can see, we are forced to. Here is why: if
>we suppose not, then (a) your last 5 minutes of conscious
>experience is equivalent to the right program going through
>a few billion states on the cpu or biological organism of
>your choice. (So far few on extropians would disagree.)

I can agree with this.

>Next, that is equivalent, in turn, to a big lookup table.
>(see my post yesterday of "going on a hike with Eugene
>Leitl" for a complete description). That is, no information
>flows between states! They are not linked by a cause and
>effect chain!

This is incorrect, as all you've done is describe a finite state
machine. You are only considering the lookup table as being a giant
indexed data repository in the way a dictionary is, without any concept of
context in the way a finite state machine has context. A lookup table that
is an equivalent representation of a program contains entries that are data
and/or code (the distinction being fuzzy) and all finite state machine
programs are reducible to a lookup table. In the pathological "pure
lookup" case, each lookup changes the state of the machine which causes it
to do yet another lookup; in practice, most lookup tables are logically
reducible to much more compact algorithms, but the concept is still
valid. To reject the validity of a lookup model is to reject the
possibility that the brain is a finite state machine.

>At each second of the hike, or of your own
>last 5 minutes conscious experience, we merely form a 30
>bit address made from (i) 10 bits of input that you are
>getting at the current millisecond (ii) 20 bits specifying
>the state that you are in.

The addressing is incorrect. You've confused the number of clock cycles
executed during the last five minutes (which is small) with the
astronomical number of possible states. You could create a hash table of
all the states used in the last five minutes and discard the rest (so that
you can meet your bit depth requirement), but then you would have a model
that was perfectly useless.

>Next, the billions of states making up your last 5 minutes
>being looked up in sequence is equivalent to them being
>looked up in any order. What difference does it make if
>they are rearranged?

Again, these billion states aren't random, each state that happens next is
a consequence of the previous state. Without this context (which any
finite state machine can provide), you have no
intelligence/consciousness/whatever. So it all boils down to the program.

>Next, this is exactly equivalent, as I think Robert
>Bradbury was getting at earlier, to having all the
>billions of state simply encoded in one long string.
>Then I ask, why do they actually have to be fetched?

Because you have to run the program at least once to know what that string
would look like (shades of the Halting Problem). And even then, if you
reduced the program to that string, it would only ever be capable of doing
the exact same thing over and over.

>And this is exactly where Egan finally went, with the
>ludicrous Theory of Dust. Unless you balk at SOME step
>above---I balk that lookups USED EXTENSIVELY AND SOLELY
>are computation, at least the kind of computation that
>supports consciousness---they you are led directly to
>the ludicrous Theory of Dust.

The problem with the Theory of Dust is that there is no shared context. It
would be the same as taking a binary program, chopping it into three parts,
running each part on three different machines at the same time, and somehow
having the software "work". All the pieces are there, all running on
computers, but the best you can hope for is a core dump.

>If I and those that agree with me are right, then this
>is a marvelous thing: somehow (a) the cause and effect,
>(b) the information flow across time, are essential for
>calculations to be conscious. I know it sounds crazy,
>but for more than ten years I've seen utterly no escape.

All you are saying is that consciousness must be able to run on a finite
state machine, which I agree with. However, from there you take the
concept to places that don't make any sense. You can't logically say that
consciousness must run on a finite state machine AND have lookup tables not
be able to produce consciousness. This is the same as saying that
equivalent programs are not equivalent where consciousness is concerned,
which I have a very hard time swallowing (a=b and a=c, but b!=c?).

There is another way.

-James Rogers

This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:59:43 MDT