Re: No AI for Nano/No Nano for copyloads

Date: Thu Jul 13 2000 - 10:41:38 MDT

Robert J. Bradbury wrote:
> You can imagine a destructive readout approach with nothing more than extremely
> very fine tissue slicers combined with near field microscopy ... In my mind this
> isn't an "upload" but a "copyload". ... a lot of antibodies or other
> neurotransmitter identifiction ... probably beyond current technology ... It might
> be feasible in ten or so years ... My guess though is that that type of readout
> would be quite slow and very expensive. ... 5 Terabytes of information just to
> store the brain map ... Getting it into main memory ... circa 2010-2015. Even then
> it is going to be a pretty sizable computer.

There are several additional problems that have to be solved. The first
is to have a good, nearly complete understanding of the dynamics of the
brain. Just knowing the shapes and positions of the neurons won't do you
any good unless you know how the operate and how they change over time.
We are still very far from understanding this area.

The second is the ethical problem of making sure it will work before
killing a human being to try it. This means that you will have to
do work with animals first, and show that they are able to operate in
uploaded form, and are not in distress.

Unfortunately, with animals it will be more difficult to come up with
good tests for how well they function, until we are able to provide them
with a relatively realistic simulated body and world to interact with.
We can't just ask, are you okay. We have to watch the uploaded rats
running simulated mazes, eating simulated food, and so on.

As a result, it will be necessary to have a relatively mature simulation
capability that goes beyond just running the brain. The simulation will
have to include the body and the environment, and will have to be complete
enough that an animal can adapt to whatever differences are encountered,
which means that it probably has to be pretty close to reality.

Also, Billy did some calculations a while back, and a low level neural
simulation (as you'd presumably require, working with raw brain slices)
requires more computing power than will be feasibly available in a
non-nanotech world.

For these reasons I think there is a lot more to the problem than just
being able to read out the brain and fit it into memory.


This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:34:33 MDT