[extropy-chat] Clock rate or rather communication delays

[Keith Henson]

At 09:01 PM 5/5/2006 +1000, you wrote:
>On Thu, 2006-05-04 at 17:10 -0400, Keith Henson wrote:
> > .
> > I think my contribution to that thread was a criticism that I have never
> > seen answered.  Namely that beyond a certain point, you get less from more
> > since the amount of computation you can do goes up with the cube, but the
> > clock rate has to go down because of speed of light delays.
> >
> > Beyond a size that isn't a lot larger than a human head, you are going to
> > get a society of minds or one that thinks very slowly.  But I have never
> > seen an analysis with numbers in it.
>
>Why would synchronicity of the clock determine the process identity?
>Chips already in commercial use have section-local clocks (P4 ALU runs
>at twice the speed of the rest of the chip) and there's design for
>clockless chips.  Would a mind operating on those chips not be one mind
>but a society of minds?

I have engaged in this discussion before and completely failed to convey 
what I consider to be the problem.  It is probably tied up in my notion of 
"spirit," that is an entity you can interact with, time subjectivity and a 
grim understanding of engineering fundamentals that nanotech will *not* change.

Assuming you can't get around the speed of light, imagine an AI mind spread 
out over a cubic light year.  If a person operating at human speeds tries 
to interact with such a creature, they better have life extension because 
there is no way such a thing can respond to them quickly in a way that 
engages the far corners of its "mind."

The same kind of problem shows up in the design of computer 
systems.  That's why looking out from the CPU we have levels of cache 
memory (extremely fast), main memory  (fast) and disk (really slow).  The 
problems have become *worse,* not better, as CPU speeds go up.

At some point depending on the maximum practical clock rate, far away 
memory or processing power becomes of low value for real time 
interactions.  Sure there are special uses (SETI, folding) being made of 
processing power on distributed net connected computers, but can you 
imagine trying to implement an AI you could talk to that way?

To put scaling numbers on this, consider a human mind  as operating at 
about 1 TPS (thought per second), supported by processors that run at 100 
Hz.  Speed of light says it could be spread over perhaps .001 light second, 
huge, a million feet or 200 miles.

Run the processor speed up to 100 GHz though and that size drops by 10 exp 
-9.

At some point in the run up, you reach the point where there just isn't 
enough volume to stick in the parts, power the hardware and cool it.

*As a guess* this is within an order of magnitude of 1 foot, which implies 
a speed up in thinking on the order of a million times.

I seem to remember that Eric Drexler came to similar conclusions in EoC.  I 
don't have time to search either my hard copy or on line, but maybe some 
reader could do that.

We don't consider the speed of light often because for our thinking rate 
and planetary dimensions it close enough to zero delay. But for creatures 
thinking a million times as fast, it is going to be a problem.

Keith Henson