Re: Don't Forget To Upload The ENS

From: Anders Sandberg (asa@nada.kth.se)
Date: Mon Jul 23 2001 - 02:11:02 MDT


On Sun, Jul 22, 2001 at 03:38:01PM -0700, J. R. Molloy wrote:
> From: "Anders Sandberg" <asa@nada.kth.se>
> > I can concieve of a very non-godlike automated system for uploading. You
> > need a scanner (lets say a nanotech version of the microtome slicer with
> > camera, giving detailled cell strcture data), an automated
> > reconstruction modeller that simply takes the data and does a fairly
> > simple conversion to a compartment model of the neurons and a runtime
> > system running the simulation. No need for even human-level AI anywhere,
> > although the pattern recognition aspect could likely be improved by some
> > extra smarts.
>
> Sure, and I can conceive of a magic potion to do the same thing. ©¿©¬
> None of the versions of hardware you mention actually exists. The "runtime
> system" you imagine would need to exceed the capacity of all extant hardware,
> and if the intelligence necessary to do this does not exceed human-level, we'd
> have one _now_.

Get real. We cannot build an orbital tower made of carbon nanotubes at
present, but that does not imply that the intelligence needed to create
this hard engineering feat is beyond current human intelligence - the
problem is simply a lack of certain material technologies. The fact
that the hardware to run an upload is extreme does not imply that you
need superintelligence to build it. The same could just as well be said
for the pentium processor.

I think you do not understand what computational neuroscientists are
doing right now. The hard part of the above scheme is not the runtime
system, we have the essentials in the form of compartment simulations
and chemical models already - they can in principle be scaled up, and do
not contain any smarts except solving a lot of coupled differential
equations numerically. Understanding the brain enough to create this
kind of model does not require superintelligence, just a lot of
dilligent research to dig up data like ionic conductances and time
constants of all ion channels of a particular type of neuron.
Intelligence helps, but it is the collective intelligence of the
neuroscientific community rather than any part of the uploading system
per se. The pattern-recognition system is the part that would require
the most smarts, since it needs to be able to determine the 3D
morphology of membranes and convert that into a compartment model. But
there are already software doing something like this, and this is likely
an area where a brute force approch could work.

> It doesn't help to underestimate the difficulty of building hardware, but I
> sympathize with your exuberant optimism, it's just that I think it makes more
> sense to build a UIM than to dream about uploading without one.

Sure, build one and I will believe you. But your own criticism of
uploading applies to UIMs too, and since you most likely clearly see why
it is wrong in that case, I hope you can extend the analogy to uploading
too.

-- 
-----------------------------------------------------------------------
Anders Sandberg                                      Towards Ascension!
asa@nada.kth.se                            http://www.nada.kth.se/~asa/
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y


This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:09:02 MST