RE: Why the Spike did not happen

From: Michael Wilson (
Date: Mon Dec 05 2005 - 10:51:47 MST

Rik van Riel wrote:
> The singularity depends on improved software, not on
> improved hardware.

I don't think this is correct in the sense that you appear
to be implying. It is certainly true that we're going to
need much better software than we have right now to get a
Singularity started, be it by designed AGI, brain simulation
or human-computer interfacing (though there is a slight
possibility that we might genetically engineer more
intelligent humans first, or make an AGI using current GAs
running on extremely powerful quantum and/or nanoassembled
hardware). But even if we can do no more than match the
computational efficiency of the human brain, even /current/
hardware can run ten million times faster and will soon be
able to scale to even higher levels of parallelism. The
physical limits of computation, and even the limits of what
looks achievable given current hardware engineering
knowledge, are vastly beyond what the human brain appears
to be capable of.

Remember that Vinge's original usage of 'Singularity'
referred only to a sharp predictive barrier beyond which we
cannot make any useful predictions about what the world
will be like. As Eliezer pointed out in 'Staring into the
Singularity', it really doesn't take much of an improvement
in intelligence (even without a change in cognitive
architecture, or the effects of being able to self-modify
at a coarse level) before this horizon pops up. The whole
'exponentially accelerating change across tens of orders of
magnitude, right up to the ultimate limits on computation'
idea, while popular in these parts, isn't a prerequisite
for a Singularity. If we can get over that initial software
problem of building something roughly as intelligent as we
are, there is already more than enough technology waiting
to take that entity/entities far beyond the realm of human
equivalence, very quickly.

As it happens I do think that there is vast scope for
software improvement as well as hardware improvement, and
that the absolute difficulty of the problem using optimal
methods scales rather slower than the power of the system
being designed, but that's just personal opinion based on
private research. I don't have to be right about this for
the above reasoning to be valid.

 * Michael Wilson


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:54 MDT