[extropy-chat] Fragmentation of computations

Eugen Leitl eugen at leitl.org
Mon Mar 26 19:16:05 UTC 2007

On Mon, Mar 26, 2007 at 06:36:10PM +0100, Russell Wallace wrote:

>    How do you test whether an entity is conscious? The most obvious
>    method is to have a conversation with it.

Very much so.

>    To have a conversation with Rock Man, you would have to build a
>    "decoding device" that actually contains all the information in Rock
>    Man's mind. It is then obvious to any reasonable person that Rock
>    Man's consciousness resides in the "decoding device", not in the rock.
>    By contrast, you can have a perfectly normal conversation with
>    Hash-Life Man anytime you want, and he will truthfully report being

Life Man, yes. Hash Life man, not so fast. From "Mind Children", pp. 157-

"But what if one wants to see the calculation in progress, as in the Newway
story? At first Gosper tried simply displaying the partial answers as they
were computed. The results were bizarre. The program advanced the simulated
time in different portions of the space at different rates. Sometimes it even
retreats in places, because some regions are described by more than one
pyramid, and the different pyramids are not computed at the same times.
A single glider advancing across the screen would cause a display where
gliders would appear and disappear in odd places almost at random, sometimes
several in view, sometimes none. Constraining the program so that it never
reversed time in any displayed cell improved things only slightly.
The best solution turned out to be not to display at all until the
calculation was finished. The pattern might start out a billion cells
on a side (necessarily mostly empty space!) and its future would be
calculated for a half-billion time-steps. The full history of the calculation
would end up compactly encoded in the hash table. A separate program could then
invovke the table entries to view the universe at any given time. The viewing
program allowed Gosper to scan, goldlike, forward and backward in time
though the evolving Life pattern. Where did that glider come from? Here it
is at time 100000. It wasn't there at 50000. Nor at 75000. Aha! This collision
just before 80000 generated it. Let's look at this step by step...

But what if Cellticks were to evolve in a hashlife space? By encoding
their universe and its evolution in such an efficient way, Gosper has played
them a dirty trick. What they perceive as the steady flow of time for the most
part does not exist. The hashlife program skips over large chunks of
spacetime without going thorugh all the tedious intermediate steps. The
Cellticks may have memories of things that never actually happened, though
they were mathematically implied from their past.

(Wow, excellent treatment. I thought the image was mine, but apparently I stole
it from Moravec).

>    fully conscious. Nor would you have doubted this in any way until some
>    engineer remarked "oh, the computer actually uses database lookup of
>    partial results instead of redoing the arithmetic wherever possible".
>    Suppose _you_ (a future upload of you) were Hash-Life Man cheerfully
>    having a conversation until the engineer came into the room and
>    remarked about the database lookup design, and the person talking to
>    you went "oh, ugh, you're half a zombie then!" how would you react?
>    Would you conclude "well I _feel_ perfectly conscious, but I guess I
>    must really be half unconscious no matter how I feel"?

How can you have a conversation with an entity which has a nonlinear
spacetime plot? The Hash Life springs to the target state, doing something
highly nonlinear in-between.

Target states are static, and nonlinear evolution doesn't make conversations
possible, unless we fall back to the degenerate state: targetting subsequent
states only.

Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
ICBM: 48.07100, 11.36820            http://www.ativel.com
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE

More information about the extropy-chat mailing list