From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Thu Mar 15 2001 - 10:41:53 MST
Ok, because this discussion is too good to miss, I have
resubscribed to the list (but if my mailbox starts
filling up with mortar shells, I'm going to figure
out how to filter that stuff out and invoice the parties
involved for my time!)
Hal said:
> This points to a problem with the idea of a simulation that covers only
> a limited period of time: how to initialize the characters who must
> inevitably be partway through their lives at the time the simulation
> begins. This is especially difficult if the characters are not zombies.
Huh? We are entering the period in time when those of us who want
to be will be mentally "on the net" via a fiber interconnect will
be able to do so. I've done the calculations on this and using
wavelength division multiplexing and a few thousand fibers, you can
definitely "tap" the brain at its full internal bandwidth. So once
this happens there is virtually no problem recreating such individuals
partway through their lives. Its the equivalent in database operations
of executing a "ROLLBACK" command. The only problem you potentially
have is the granularity of time resolution that you want to be able
to "ROLLBACK" to. But with 10^40+ bits at your disposal and the
fact that you only really need to save the "difference" information
between the ROLLBACK states (read: you can compress the ROLLBACK
data a *lot*) I suspect there are a huge number of states that you
will be able to start the simulation at.
Now, rolling back to a pre-Matrix state is going to be a bit trickier.
I suspect its going to take a detailed understanding of how the
brain really works, a reasonable approximation of the genetic code
for the individuals involved and probably a fair amount of hypothetical
"back-tracking" simulations. Since you know the historical reality
to some accuracy, you have to backtrack from the current reality
all the way through to the historic reality, then allow it to run
forward to see if you get back to the current reality. If you do then
one assumes your recreation of the historical state was accurate enough.
You run into problems due to chaos theory. You probably need to
allow it to run forward multiple times, if you end up in your
current state 99 times out of 100 then your past recreation is
"good enough". I expect it will be a very interesting branch
of science and mathematics to determine whether there are multiple
initial historical states that lead to the present state. That
will set some interesting limits on how "accurate" you need to
make the historical recreation. [On most days, it just doesn't
matter whether I get out of bed at 7:00 or 7:01... etc.]
> The fundamental paradox is that if the Simulators are not able to know
> how humans will behave without running the simulation, then they will
> not be able to come up with realistic memories and histories for adults
> at the start of the simulation. Or, putting it the other way, if they
> are able to tell what people will be like and what they will do without
> running a simulation, then they don't need to run any simulations.
I think more probably, you have to run enough simulations to get
the "standard forms" for the individuals you want to put into it.
Due to genetic and socialization constraints humans are remarkably
similar to each other (when one considers the phase space of how
they "could" be). If you look at psychology you have all of the
standard ways of "classifying" individuals. I suspect you could
could have lots of pull down menus on the simulator control panel
for this type of stuff, e.g.
Sex: ____________
... Scroll down to "male, normal"
Intelligence: ____________
... Scroll down to "very bright"
Environmental Background: _________________
... Scroll down to "strict religious bordering on oppressive,
prototype Eliezer Yudkowsky's home"
etc.
Robert
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:06:23 MST