From: Hal Finney (hal@rain.org)
Date: Wed Jul 23 1997 - 14:57:53 MDT
Mark Crosby, <crosby_m@rocketmail.com>, writes:
> Cosmology aside, this is precisely what I never understood from _The
> Physics of Immortality_: My body/mind is a product of innumerable
> interactions with my environment. How can neural nets, or whatever
> Omega might use, spring full-blow like Athena from the head of Zeus
> without some kind of interactive growth or training? How can
> ~Qbrute-force resurrection~R create something human-level or greater
> *that once existed* ~Qfrom scratch~R without also simulating all the
> environmental input factors that conditioned that being?
Generally, it does seem correct that in order to create a person who
thinks he has experienced X, we have to have simulated X to at least
the level of detail remembered by that person. If you want to create
someone with memories of combat in Vietnam, you will have to have created
specific combat events for him to remember.
It may not be necessary though to simulate every detail of the Vietnam
experience down to the water droplets dripping from leaves in the jungle
all around him. According to some studies we've discussed here, the
actual amount of data which goes into long term memory is very small on
average, perhaps only one bit per second or so. Naturally, some times
are remembered much better than others, and even forgotten events can
have a significant influence on future behavior. But broadly speaking it
may be possible to have a way to create a person with memories which
realistically simulate a Vietnam experience without actually simulating
every detail.
Even if it turns out that detailed simulation is sometimes necessary,
there can still be economies of scale. There could be a generic
20th century simulation which could be used as the basis for creating
huge numbers of possible 20th-century citizens.
> Even if there are only x^y states of ~Qmatter~R possible and infinite
> computational resources at Omega could compute all possible
> combinations, how could ~QThe System~R ever *select* those
> configurations that might have had some resemblance to actual history
> (which would seem to be required in order to call this a
> ~Qresurrection~R)?
This reminds me of an interesting point raised by Hans Moravec here on
this list once. As Mark says, in some ways it seems that selection is an
important step in giving "reality" to the simulated systems. If you had
infinite computer power, you could simulate every possible system in every
possible sequence of states. But this raises the question of whether
you have actually accomplished anything in doing so. In some sense,
it seems like by generating every possible system, without actually
selecting out those which are meaningful, you have blurred them all
together and it is as though you haven't generated any systems.
I'm not sure I agree with this; Moravec is a Platonist and believes that
all these systems have reality independent of whether they are actually
instantiated. Simulating a hypothetical being merely gives us access to
him; he exists independent of our efforts.
A more materialistic view is that the only systems which can have reality,
including awareness, are those which are actually implemented at some
point in the real world. In that case, simulating a system which
represented a conscious being would in fact bring him into existence.
This would be true even if his simulation was just one variant on an
infinite number of themes, monkeys typing on infinite typewriters.
Hal
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:44:38 MST