Re: Singularity and AIs -software

From: Robert J. Bradbury (
Date: Mon Jan 28 2002 - 16:23:02 MST

On Tue, 29 Jan 2002, John Casey wrote:

> Gregory Bateson used to say, I think, that intelligence basically was the
> ability to find the pattern that connects two objects or events. I'm curious
> why we've had so little luck devising programs that can do this.

William Calvin would probably argue that it derives from the
interaction of two thought patterns overlapping one another
in the brain.

> And to be sure, the weight of the evidence,
> at least as far as modern technology is concerned, would seem to suggest
> that western cultures have the jump on innovation.

Be careful -- there is probably a selection effect in that "Westerners"
may have been higher levels of novelty-seeking genes. We discussed
this on the list a couple of months ago.

> But. If intelligence includes innovation, then innovation also would seem to
> involve generating and testing "patterns which connect."

The problem is -- do they "connect" from a common sense perspective
of being innovative or useful or are they simply "interesting"?
The connection of "sad--cow" generates an entirely different
pattern in my brain than the two patterns generated by "mad--cow"
which is different from the somewhat fuzzy "cow--mad".

> It ought, therefore, to be possible to construct an algorithm for
> pattern generation and testing, and thereby to create a bona fide intelligent machine. I don't
> think I'm on new ground here, at least not on the "is it possible?"
> proposition. Maybe a bit more so on the question of why we ain't done it
> yet.

Its "hard" but we are getting closer. See some of the discussion
(and refs) in my Nano@Home paper:

As someone's sig says, "everything is easy once you know how to do it".
It is questionable whether humans could ever properly wire together
100 million transistors -- but there are computers that can accomplish
that clever trick. What one needs are the metrics and heuristics
that generate useful patterns and efficient means for recognizing
good combinations in specific knowledge domains. We are slowly
(very slowly in most cases) managing to accomplish that.


This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:37 MST