[extropy-chat] Moore != AI

[Dirk Bruere]

Eugen Leitl wrote:

>On Thu, Mar 31, 2005 at 02:57:03PM +0100, Dirk Bruere wrote:
>
>  
>
>>Well, we are already at 10^14 FLOPS and at a guess a FLOP takes more 
>>    
>>
>
>Is it affordable? Is peak flops a meaningful measure for neuronal code?
>
>  
>
It's an indicator of processing capacity, given that a FLOP is a rather 
complex operation.

>>processing than a dendrite/axon. By 2015 we may well be up to 10^17 
>>    
>>
>
>You're guessing. You're wrong. All practical packages burn up OPS like
>there's no tomorrow.
>  
>
I also suspect that neural net simulators with dedicated h/w will not 
burn many FLOPS at all. Maybe integer arithmetic is all that is needed.

>Getting an O(1) scaling with problem sized (if matched by hardware) is in
>theory easy (spatial tesselation CA over a 3d torus of nodes is an example),
>but no such practical codes for AI exist. Not yet.
>
>Software engineering doesn't follow Moore, you'll observe.
>
>  
>
How do you measure that then?
It certainly seems to fill available memory and consume available power 
in line with Moore's Law.

>>FLOPS in top of the range supercomputers.
>>Neural sim does assume specialist h/w though.
>>With such simulating 10^10 neurons each with 10^4 axons doesn't seem all 
>>that difficult from a computational POV.
>>
>>    
>>
>>>Even if we knew, how to extract the relevant information (we don't), and 
>>>had
>>>a system able to simulate the relevant aspects (we don't).
>>>
>>>
>>>
>>>      
>>>
>>I'm a bit more optimistic than that.
>>    
>>
>
>Do you have a rational reason for optimism?  Morphogenetic part of genetic
>networks is currently a complete unknown. Scanning is currently limited to
>TEM -- properties of the shape are complete guesswork. Some simple systems
>are solved, a generic wet neuro simulator is quite beyond the state of the
>art.
>
>  
>
http://www.newscientist.com/article.ns?id=dn3488

>Have you seen how e.g. Neuron scales?
>
>  
>
>>I would be suprised if more than 10% of the brain was actually involved 
>>in interesting stuff, such as consciousness.
>>    
>>
>
>I don't know what consciousness is. I do know that you don't know how to tell
>which aspects of the scanned system are relevant, and which are omissible.
>
>Not trying to put you down; nobody else does. Yet.
>  
>
Yet it *is* known how the brain is divided into areas of rough 
functionality, and many of those are almost certainly not needed by a 
sim AI.

-- 
Dirk

The Consensus:-
The political party for the new millenium
http://www.theconsensus.org



-- 
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.8.6 - Release Date: 30/03/2005