X-Message-Number: 27329
Date: Sat, 5 Nov 2005 12:02:43 +0100
From: Eugen Leitl <>
Subject: Re: [CN] to the mouse ... and beyond
References: <>

On Fri, Nov 04, 2005 at 11:37:45AM -0500, Brent Thomas wrote:
> As Thomas pointed out 2.64 million atoms is small potatoes in the context of
> things...but as a measure of what is possible today, and what will become

It is not small potatoes at all, considered the amount of
massively parallel computation required for an atomically
accurate simulation this size.

You can assume that a segment of a dendritic tree needs
about as much computation as an atom in MD, but ms for fs. 
With Avogadro amount of switches there is enough resources 
to directly tracing out biological structures as virtual 
circuitry, nevermind more efficient packet-based approaches.

> possible exponentially soon thereafter it is a great indicator of
> mind-boggling levels of complexity soon to follow. I thought this particular
> example was interesting because it was an example of a researcher performing
> a 'bottom up' atomic level simulation...and if you can simulate that with
> enough precision you can effectively simulate functionally anything in the
> world (including a brain) without having to understand it beyond the
> 'snap-shot-in-time' capability of an atomic level scan. 

Atomically accurate models of individual assemblies (e.g. an ion channel)
are a good source of parameters for higher-order models, which are many
orders of magnitude less wasteful.
 
> I do think it will take us much more effort to understand and optimize the
> workings of a full brain simulation beyond the atomic level because that
> implies we actually understand what we are doing not just performing a copy
> (and understanding is always harder than imitation).

Atomically accurate models of individual people are fundamentally feasible,
but utterly impractical. Crunching a cubic micron at atomic detail in us-ms
range is something different, and highly worthwhile.
 
> However the real point I wanted to make was about how neat this particular
> example is as an indication of what some particular researcher was able to
> accomplish with today's technology (6x larger than any previous simulation)
> and to plant the seeds of thought in the archives and our reader's minds as
> to what may someday become possible.


10^9 atom simulations (simple metals, admittedly) were done about end 1990s, 
IIRC.
That particular simulation didn't use the full resources of this particular
machine, either.
 
> If you share an interest in this (and as readers of THIS list you probably
> do and are probably already familiar with the concepts) you might find this
> interesting
> http://72.14.207.104/search?q=cache:y0ZZUXxdF0YJ:www.pivot.net/~jpierce/star
> ing_into_the_singularity.htm+moravec+transfer&hl=en
> 
> Particularly the references to moravec transfers. 

Don't forget computationally studying what vitrification does to biological 
assemblies, specifically the amount of irreversible information erasure,
and assessing how much of it is tolerable for individually accurate person
reconstruction.

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

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=27329