Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornell!travelers.mail.cornell.edu!news.kei.com!news.mathworks.com!uunet!zib-berlin.de!news.belwue.de!news.uni-stuttgart.de!rz.uni-karlsruhe.de!stepsun.uni-kl.de!uklirb.informatik.uni-kl.de!schmid
From: schmid@informatik.uni-kl.de (Klaus Schmid)
Subject: AI can't incorporate adaption
Message-ID: <1995Feb16.194733@informatik.uni-kl.de>
Sender: news@uklirb.informatik.uni-kl.de (Unix-News-System)
Nntp-Posting-Host: skolem.informatik.uni-kl.de
Organization: University of Kaiserslautern
Date: Thu, 16 Feb 1995 18:47:33 GMT
Lines: 30


> One reason to dismiss classical (say computational/symbolic/etc.) 
> AI models of cognition is the fact that they cannot incorporate
> adaptive processes as evolution, development, learning
> (all this is possible with biologically motivated models
> like neural networks).

First a few questions regarding the words you use. What do you mean by
a computational model -- In what sense are neural networks (especially
artificial NN) NOT computational??
What do you mean with learning in this context? What can an artificial 
neural net do, that can't be done by a symbolic machine learning system?

Indeed in the comparisons I know of, even (simple) machine learning 
algorithms like ID3 did clearly outperform Neural Nets (e.g. back-
propagation nets) on the same task.
This is not only the case wrt efficiency of the learning (which 
is rather trivial ;-)  ), but also with respect to the quality of
the results of learning. (This has been the case with the learning
in the context of numerical data; a situation primed in favor of 
neural nets.)

Besides, have you ever tried to make a theorem prover (or any other 
symbolic system) learn using neural nets?


CU
Klaus


