RE: Hardware Progress

From: Ben Goertzel (ben@webmind.com)
Date: Sat Jun 23 2001 - 15:40:20 MDT


>
> Hardware is *not* the problem. Good theory & software design are.
>

While I strongly share this *intuition*, Peter, I don't think that you or I
have the data to make this pronouncement in such a definitive way.

The statement I would 100% support is: "Given the current state of our AI
theory and software design, it DEFINITELY cannot be claimed that hardware is
our obstacle to creating real AI."

With the Webmind AI design, which I believe is the most fully-developed one
on the planet (Peter's and Eli's included), there is *still* no 100% solid
way of telling how much processor power will be needed to make Webmind how
smart.

We do know that the current C linked-list-based implementation is vastly
more efficient than the previous Java agents-based implementation. And we
can make assessments as to the scalability of various system components
(inference, concept formation, pattern recognition, etc.). But the emergent
behavior of the whole system remains an unknown, except on a *qualitative*
level.

Perhaps there are other AI designs that are more quantitatively,
analytically tractable than Webmind. But, I very strongly doubt it; I
suspect that this analytical intractability is an inevitable property of any
system complex enough to give rise to the phenomenon called "mind."

Once we have built some intelligent AI systems, and experimented with them,
we'll start to create a quantitative science of mind. Then we'll know, in
retrospect, exactly which year it was first viable to create a digital mind
using available hardware.

At this point, we're just ignorant fools in the science of mind, and so we
have to go on intuition, supporting our intuitions with facts insofar as
possible.

-- Ben



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT