X-Message-Number: 25516
Date: Tue, 11 Jan 2005 10:26:02 -0500
From: Jeffrey Soreff <>
Subject: The Singularity Is A Fantasy

>Existing connectionist computing paradigms have universally failed to 
>scale. No one has been able to demonstrate non-biological neural 
>networks that reproduce the behaviors of even the simplest animals. 

Ok, I've seen the hype for neural networks years ago, but
I'll take your word for them running into a scaling wall.
A pity that failures of hyped technologies don't get more
widely publicized.

>Neural Networks are notoriously vulnerable to combinatorial complexity 
>and no one has demonstrated a paradigm wherein the intelligence of one 
>NN can be combined with the intelligence of another to any constructive 
>purpose. And genetic/evolutionary computing is yet another Turing/Von 
>Neumann paradigm technology, bound by the complexity constraints of all 
>T/VN computing - combinatorial explosion affects it the same as the 
>other AI games.

>Look, a lot of very bright folk, including a number of cryonet 
>luminaries, have spent a lot of time bashing their heads at AI. What 
>they got for their trouble are a huge number of very 
>real-world-expensive problems that they can't solve because of their 
>inability to deal with computational complexity at full scale. From 
>routing planes and trains through catching credit-card fraud and 
>insider trades, from reconciling financial databases - so you DON'T get 
>seven different credit card offers from the same bank - through 
>figuring out drug interactions ... the industrialized world is lousy 
>with unsolved problems of combinatorial scale.

>These are known in the abstract as NP-Hard problems. The commonality to 
>them is that you can mathematically prove you can't solve them in 
>polynomial time on T/VN computers. This is to say, no matter how big 
>your T/VN computer, these problems rapidly grow to the point that it 
>can't cope. Now it may be that orienting a nanobot or coordinating the 
>activities of a millionty billionty nanobots may not entail NP-Hard 
>problems. But that would be extremely unlikely - in every other problem 
>domain the bloody things are everywhere. They are the major constraint 
>on the problems we can attempt to solve with a computer - which is to 
>say they are the major constraint on modern technology, period.

Yes, NP-hard problems are almost certainly exponential (though,
as of the last I'd heard, NP!=P still hasn't actually been _proved_).
That said, the bulk of NP-hard problems have some sort of
approximation which is _not_ exponential.  I work in CAD, and we
routinely compute approximate solutions to many NP-hard problems.
For instance, a routing problem that has to go through many points
in a large area can be approximately broken up into independent
problems in chunks of the large area, if one accepts the penalty
that the solution is forced to go through imperfectly selected
points where the chunks meet.  You don't get the true optimum this
way, but you can force the run time down to a polynomial of the
problem size.

My personal view is that a singularity is unlikely, but for a
couple of different reasons:
   1) As Robin Hanson has pointed out, there are many different
      potential bottlenecks in an economy.  If we HAD human-equivalent
      AI and infinitely fast CPUs, there are still physical bottlenecks
      tn how fast materials can be moved around, how fast physical
      experiments can be performed, etc.
   2) Even if we had infinitely fast CPUs, we don't KNOW how to build
      human-equivalent AI.  As far as I know, no one has been able to
      do "common sense" reasoning robustly, at any speed.  A lot of
      classical AI was phrased as search problems, and a lot of search
      problems have the combinational complexity problems you describe,
      but I haven't seen experimental evidence that any of the existing
      AI paradigms, including search, can do common sense reasoning at
      all, even with unbounded computing hardware.  Is there experimental
      evidence of this?
My _guess_ is that once hardware substantially faster that the full
bandwidth of a human brain is routinely available, AI researchers
will probably find ideas which do permit human-equivalent AI, but we
aren't there yet, and I have no idea how long this might take.

   Best wishes,
   -Jeff

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=25516