From: Aaron Davidson (ajd@ualberta.ca)
Date: Sat Sep 11 1999 - 19:53:55 MDT
Gregory Hather <student@calpoly.edu> wrote:
>Some 50 years after Turning, look where IA is now. It stalled because
>it is to hard to program an adult human like mind. Instead, scientists
>should have persued artificial neural networks so the system could learn
>on its own and generalize by modifiying the strengths of its simulated
>conections. Rather than faking intelegence, this system would learn
>like a human. ANS is definatly the future of comuter intelegence.
First of all, IA is not AI, which is what you mean.
Secondly, while ANN's are cool, they are still very difficult to work with.
You can't just slap together a huge neural net, wire it up to some
input/output, switch the sucker on and start breast feeding it, shaking
rattles and reading it childrens books -- and expect it to start learning
and growing into an intelligent being. How the network is organized, the
ways in which it is connected, the learning algorithms used, and a million
other things that control how the NN learns and develops. When we are born,
we already have all this stuff built in. The brain has many predefined
areas with specific functions, and the way it learns and develops has been
tweaked over millions of years of evolution.
Thirdly, why do you believe that a symbolic or classic AI is 'faking
intelligence'? I don't want to start another debate on what constitutes
intelligence, but why should we be so naive to assume that the human way is
the only way to be intelligent?
I say if it walks like a duck, quacks like a duck, writes a book on
metaphysics, and can teach you advanced calculus, it's intelligent.
AI researchers have been attempting what you suggest both with ANN's and
symbolic systems for decades now. It just happens to be a very difficult
thing to do.
--Aaron
+-------------------------------------------------------------------------+
| Aaron Davidson <ajd@ualberta.ca> http://ugweb.cs.ualberta.ca/~davidson/ |
+-------------------------------------------------------------------------+
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:05:07 MST