[extropy-chat] Dark matter and ET

Tue Jul 19 20:41:03 UTC 2005

On Tue, Jul 19, 2005 at 11:55:25AM -0700, Damien Sullivan wrote:

> > Computronium is a molecular crystal made from computing cells, optimized 
> > for doing computation. DNA is nothing like computronium. Dry cold DNA is a

I. Amato (Science, New Series, Volume 253, Issue 5022 (Aug. 23, 1991), 856-857
http://leitl.org/amato.pdf writes "Thus the researchers claim half-seriously that they 
are creating a versatile new form of matter which they call programmable matter or
computronium. "In programmable matter, the same cubic meter of machinery can become
a wind tunnel at one moment, a polymer soup at the next; it can model a sea
of fermions [elementary particles], a genetic pool, or an epidemiology experiment
at the flick of a console key", "Margolus shares a similiar vision: a cellular
automaton computer with components on the atomic or molecular scale. ""Such a machine
would be a kind of a computing crystal with all parts participating in the
computation," Margolus says -- a lump of "pure computronium."". (Also see
http://people.csail.mit.edu/nhm/cc.pdf for more on Margolus' view).

> Is a rod-logic computer a crystal?  I've always defined computronium as matter

A cellular system made from unit cells made from NEMS elements (whether buckling
bucky, or diamond rod logic, or spintronics, it doesn't really matter) certainly qualifies.

> optimized for computation; exactly how will depend on the tech available.  And
> as Wikipedia on computronium points out, on the goals: speed, storage, energy
> efficiency, mass efficiency, cost...

That article needs fixing. Wikipedia has some gaping holes still, particularly
on lunatic fringe material as this.

> > No, in terms of energy efficiency and functionality concentration a human or
> > an insect brain is very far removed from what we can currently do. 
> 
> This seems in contradiction of the facts I possess, such as our difficulty in
> matching insect functionality, with devices which weigh more and consume more
> power.

Nonono, you misunderstand. I agree with you absolutely. Our systems suck in
comparison to even the humble bumblebee, nevermind the human primate.

> > > But I figure a brain has 1e6 to 1e8 times the raw computing power of a
> > > desktop CPU (or more accurately, would take that many CPUs to be
> > > emulated), while
> > 
> > We don't know what the equivalent is. There are no comparable benchmarks.
> 
> We know the brain has 1e11 units, 1e14 synapses, and update frequencies of up
> to 1000 Hz.  1e17 flops.  I can see needing 'only' 1e14 in practice, if much

I've been guilty of similiar estimates in the past, but have grown wary of them
since. They're misleading at at least two levels: the description of the physical
system, and at how much computation work that system is equivalent to. Also,
people automatically take peak Flopses of large current iron, and invoke Moore,
and all kind of breezy Kurzweilian handwaving ensues.

> of the brain is idle at any moment and 100 Hz is more like the real update
> speed.  And about 1e14 bytes of RAM.

Computation in the dendritic tree breaks the assumption that you only have to
deal with the neuron body and the synapse, spikes are more about timing than
the frequency, and even with relative addressing there's no way to code 14e14
synapses with 1e14 bytes.

Given that we'll be seeing Avogadro number of bits in individual installations 
within the next 50 years, or so, it doesn't really matter, though. 

> > CPUs alone are useless. You need to look at the total mass of a Blue Rack,
> > included the machine room and the air conditioning versus a ~75 kg human.
> > Here the efficiency becomes particularly abysmal.
> 
> I don't see it.  100 W for the total human system, with advanced face
> recognition, speech recognition, speech production, planning, robotics, Go

Yes, it is the power efficiency and integration density that is particularly 
amazing. Blue Gene takes 28.14 kW/rack, at 64 racks (0.3 PFlop (mythical) peak) it's
about 1.8 GW total, while falling very short of a human primate even if using bogus 
Flops equivalent estimates. 

> playing (potential) and partial protection against hostile replicating goo, to
> put a few things in CS terms.  Well, for the naked human; if you start
> counting First World lifestyle it's 1e4 Watts.  But brains being practical
> computronium doesn't have to mean using human beings; it can mean grown arrays
> of neurons, or cyborged animals.

I'm expecting small assemblies of molecular electronics completely vanquish
biology in terms of density, power envelope and speed in about 20-30 years.
Scaling this up will take some time more, but that it will happen within our
lifetime (barring major breakdowns of the world order) is pretty likely.  

-- 
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
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20050719/6590157d/attachment.bin>