Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornell!travelers.mail.cornell.edu!news.kei.com!news.mathworks.com!news.alpha.net!uwm.edu!vixen.cso.uiuc.edu!howland.reston.ans.net!Germany.EU.net!netmbx.de!zrz.TU-Berlin.DE!cs.tu-berlin.de!fu-berlin.de!news.dfn.de!RRZ.Uni-Koeln.DE!news.rhrz.uni-bonn.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: Re: AI can't incorporate adaption
Message-ID: <1995Feb22.181719@informatik.uni-kl.de>
Sender: news@uklirb.informatik.uni-kl.de (Unix-News-System)
Nntp-Posting-Host: gentzen.informatik.uni-kl.de
Organization: University of Kaiserslautern
References: <1995Feb16.194733@informatik.uni-kl.de> <3ia1g5$kc7@sunserver.lrz-muenchen.de>
Date: Wed, 22 Feb 1995 17:17:19 GMT
Lines: 64

In article <3ia1g5$kc7@sunserver.lrz-muenchen.de>, t383102@hp22.lrz-muenchen.de (Marc D. Weitze) writes:
|> Of course, biologically motivated models like ANNs are computational
|> in the sense that they run on digital computers. But this is peripheral,
|> for computers are only a tool in ANN modeling. The point is that ANNs
|> can "show how a system might convert a meaningful input into a meaningful
|> output without any rules, principles, inferences, or other sorts of
|> meaningful phenomena in between." (Searle, 'Rediscovery' (1992), p. 246)
|> This stands in contrast to computational models.
There are rules, principles, inferences in NN - or how would you call the 
learning rules the (programmed) ANN adheres to: Obviously they are rules in the
same sense in which there exist rules (e.g.) in ID3 for selecting an attribute
to test on.
Besides this (so-called) classical learning algorithms are not subject
to rules either. Their behaviour is dominated by the examples (in the sense
this is true for NN ;-)

|> Biologically motivated ANNs lend themselves to adaptation, because 
|> adaptation is a biological category: Learning (changing synapses), development,
|> and evolution (GAs) is easily transferable from biology to ANNs. This
|> is not the case for classical GOFAI architectures: Adaptation does not
|> play an 'intrinsic' role in computation/logic/language/etc. and cannot
|> be incorporated in those models free and easy.
- What is a GOFAI architecture ?
Here you are right - from a psychological point of view. I do think that the
important criterion is, what performance the different architectures show at a
learning task.

|> Bobrow and Winograd found a nice metaphor in 1977: 
|> "Current systems, even the best ones, often resemble a house of cards...
|> The result is an extremly fragile structure which may reach impressive
|> heights, but collapses immediately if swayed in the slightest from 
|> the specific domain (often even the specific example) for which it
|> was built (Cognitive Science, Vol.1, p. 4).
Indeed it is a NICE little metaphor, however, it goes besides the point. 
In 1977 none of the (so-called) classical learning algorithms for learning
from examples existed. So obviously, they could not address this metaphor to 
these systems.
Additionally, these algorithms have been used with success on a wide range of
tasks, so what is said by this metaphor is obviously not correct for these 
systems.
I have the impression that this metaphor was originally addressed to the expert
systems of this time. For these it was true - as far as I know.

|> It is worth further examination if it be correct to label algorithmic
|> processes (e.g. in ID3) as 'learning'.
My point of view is that one can measure learning only by the I/O-behaviour
of the system.
Then because the (so-called) classical algorithms produce similar behaviour 
to this of ANN, consequently they learn, too.

However, if you like to decide this point based on the implementation. And
decide that anything algorithmic is not learning, then you have to consider
that
  ANN are no implementation of NN because they are algorithmic and NN are
  not algorithmic (because they learn and anything learning can not be
  algorithmic.) 


|> Best
|> Marc-Denis Weitze.


Regards 
	Klaus
