From: Robert Bradbury (bradbury@genebee.msu.su)
Date: Wed Apr 19 2000 - 08:21:02 MDT
On Tue, 18 Apr 2000, Eliezer S. Yudkowsky wrote:
> Spudboy100@aol.com wrote:
> >
> > Ok, so what factors have to occur before a "singularity " is attained? Have
> > you estimated how many Instructions per Second or how many Bitstreams?
>
> My current thinking on hardware is 1e15 floating point ops/sec, with a
> minimum speed per CPU of 1e9 ops/sec and 1e12 ops/sec preferred. Memory
> required: At least 1e12 bytes, probably more. We'll have the hardware
> before we have the software.
Eliezer, have you given any thought to the possibility that as we understand
what the architecture needs to do better, we will construct more efficient
ways of doing it? For example, the recently announced (French?) chip
(for $6) that would do most of what the human vision system does in
terms of things like object tracking. I really doubt it is close to
the optic nerve in power, but would seem to have collapsed the necessary
algorithms onto current hardware. Similarly the software that now
does voice recognition or even OCR. That stuff runs on hardware that
is probably much slower than equivalent brain areas.
If Calvin *is* correct, I would have to believe the functionality of
the # of columns in the brain can be collapsed into hardware that
requires much less processing power.
The more and more I think about it, the more I suspect the real measure
of brain capacity is in interneuron bandwidth. That is the number we
will need to match effectively.
Robert
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:28:06 MST