From: Ben Goertzel (ben@goertzel.org)
Date: Mon Apr 01 2002 - 07:34:52 MST
I think that there *may* be enough computing power out there to enable
intelligence at the human level or beyond.
However, simply implementing a standard neural net architecture in a
massively distributed fashion, will not do the trick.
Exploiting the computing power on the Net within an artificial general
intelligence (AGI) architecture is not a trivial task by any means. First
one needs the AGI design, then one needs to work out which parts of it can
be profitably implemented on a huge network of weak, unreliably connected
and unreliably available processors. We have gone fairly far down this path
with the Novamente AI system. Our conclusion is that
1) there ARE important parts of the Novamente system that can be carried out
in a Net-wide, massively distributed fasion
2) one needs a centralized core (which may be distributed, but across more
powerful and reliable and reliably connected machines) to guide and control
this massively distributed component
So, we have a moderately detailed design for 1), but we are spending our
effort at the moment on 2).
In short: Yes, those resources are valuable, but the conceptual aspects of
utilizing them are even more onerous than the technical aspects.
-- Ben G
> -----Original Message-----
> From: owner-sl4@sysopmind.com [mailto:owner-sl4@sysopmind.com]On Behalf
> Of Max Comess
> Sent: Monday, April 01, 2002 1:55 AM
> To: sl4@sysopmind.com
> Subject: Singularity sooner...?
>
>
> Hi sl4,
>
> I was wondering, how fast would a neural net over the internet
> be if it was by distributing an application to run much smaller
> neural nets
> off idle computers. Would there be sufficient connection speeds and
> stability to allow meaningful and useful communication speed
> between groups
> of neurons to enable something transhuman?
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT