From: Ben Goertzel (ben@webmind.com)
Date: Sat Jul 28 2001 - 11:46:17 MDT
> Ben,
> Perhaps a Better analogy would be instead of
> bird augmentation...dirgible development.
The dirigible analogy is interesting, but I don't find it as convincing as
my original analogy because dirigibles are not natural systems, whereas
birds are.
Actually, the airplane analogy for AI is one I've thought about a lot. As
is well known, many famous and brilliant scientists went on record with
mathematical proofs that human flight was impossible. Then these country
hicks pulled it off. The first airplanes sucked, but they worked -- even
though the theory underlying them was pretty crude, and only fully
understood decades later. It was amazing how rapidly and usefully the
theory developed after the proof of concept was there. Now of course
airplane construction follows theory rather than violating it.
I suspect the same thing will happen with AI. The so-called "AI experts" of
the academic mainstream argue why it's very difficult, there's so much we
don't understand about nonmonotonic logic, or long-term synaptic
potentiation, or whatever their particular angle is. Meanwhile someone like
me or Eli or Peter gets the first real AI to work, even though it's kinda
clunky like the Wright Brothers' first plane... and then, with the real
thing as an experimental tool and a stimulus to the mind, theory really
takes off and develops side by side with practical experimentation as in a
*real science* (which AI barely is at the moment).
The relation between natural flight (birds, insects) and airplanes is loose.
Yup, planes have wings like birds and bugs do. But they don't flap them.
There were many, many early failed attempts to make airplanes with flapping
wings. But we *still* can't do it. And we still don't fully understand the
aerodynamics of all birds and insects either -- in some cases it's still
mysterious to us how they fly as well as they do. The Wright brothers took
the wings and basic shape from birds, but added a different propulsion
system -- one of a type that never would have evolved, because it involves
wheels, gears and so forth, and evolved physical systems never seem to
involve *separated* moving parts (e.g. an axle and the hole that the axle
sticks through).
Similarly, in building AI, I don't think the best path is to augment human
brains OR even to imitate them closely. Rather, I think we need to look at
the conceptual principles underlying human brains, insofar as we understand
them, decide which of these principles we want to carry over to our
engineered system, and then with this limited set of principles, improvise
based on the theory and intuition we have at our disposal, given the
constraints imposed by the materials at hand.
Humans have a pretty strong record of creating new technologies that were
believed to be impossible: airplanes, radio, TV. ummm... spaceships...
vibrators? ;D...
We do not have a similar track record of spectacularly modifying natural
evolved systems. We have a great track record of *breeding* evolved systems
(artificial selection), but the most important cases of this (the creation
of domestic animals and crops) have taken thousands of years. Modern
breeding of animals and plants still hasn't created anything as important
as, say, wheat or rice, or the dog or horse.
Eli and I agree on this thoroughly. I love genetics and proteomics and I
think that fantastic work is going to be done in these areas, but I think
this work will before long be *enabled* by AI's that are able to assimilate
and reason about immense and diverse databases of molecular information.
All in all, I'll give you 20-1 odds it will be AI's helping humans to figure
out how to genetically engineer superhuman bio-beings, rather tha the
genetically engineered superhuman bio-beings who help to create AI.
-- Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT