re: features of an AI code

From: Yan King Yin (
Date: Tue Jul 13 2004 - 05:59:13 MDT

Eugen wrote:
> * code for today's/next year's commodity iron, but think
> about portability to massively parallel async molecular
> electronics systems with mole amounts of switches.

I think if we can have massive parallel processing, then
human-like intelligence would be fairly easy. There is no
dispute about this, but the controversy is over how to do
it. There're 3 possible ways (probably more):

1. As Eliezer suggested in LOGI, try to exploit the strengths
of von Neumann hardware (floating point math & stuff).
But you need to be very clever. Also, some problems have
been shown to be inapproximable in polynomial time.

2. DNA computing. DNA can do hybridization, which can be
seen as a sort of approximate string matching which is an
NP-hard problem. So what's the problem? I think the
problem is that there is currently no way of synthesizing de
novo DNA sequences *without* a template. If you MUST
use a template, that means you're not free to choose the
variables (which is required by the NP problem). Therefore
it boils down to a way to synthesize DNA sequences at will.
This, unfortunately, is almost as hard as molecular nanotech.

3. Silicon-based parallelism. The approach developed by
Stan Williams at HP Labs (and some others as well), basically
the cross-bar array with molecules trapped in between. But
this also has a serious problem. The silicon problem is all
about *wiring*, and any chip must have its number of pins
somewhat < 1000 (currently). That almost forces you to use
multiplexing. The result is that at any time, the number of
bits being addressed *simultaneously* is constrained by
the number of pins (~ wires).

Very serious problems ahead.

Find what you are looking for with the Lycos Yellow Pages

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:48 MDT