Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!howland.reston.ans.net!news.sprintlink.net!in1.uu.net!news1.digital.com!nntp-hub2.barrnet.net!pacbell.com!amdahl.com!amd!amd.com!txnews.amd.com!news
From: Joseph Bridgewater <bridgwtr@vanzandt.amd.com>
Subject: RFD: My Learning/Thinking Neural Network
Content-Type: text/plain; charset=us-ascii
Message-ID: <DDBsH6.BFr@txnews.amd.com>
Sender: news@txnews.amd.com
Nntp-Posting-Host: corgi
Content-Transfer-Encoding: 7bit
Organization: Advanced Micro Devices, Austin, TX, USA
References: <DCyy7A.K4B@txnews.amd.com> <408vo6$eio@seralph9.essex.ac.uk> <DD23ww.I5D@txnews.amd.com> <40lfc8$m3n@bert.traverse.com>
Mime-Version: 1.0
Date: Tue, 15 Aug 1995 00:10:17 GMT
X-Mailer: Mozilla 1.1 (Windows; U; 32bit)
Lines: 51

D Poinsett <d.poinsett@traverse.com> wrote:
	>Joseph Bridgewater <bridgwtr@vanzandt.amd.com> wrote:
	>>...
	>>If you could recommend a task to prove it's ability to think, please do. 
	>>...
	>
	>Two suggestions...
	>1) Get "IT" to recognize something too good to be true.
	>2) Get "IT" to make up a joke.
	>
	>Seriously though, good luck in your endeavors. Your freshness and 
	>enthusiasm are evident. Keep working.

I'm glad you included the word "Seriously"....although I believe those two suggestions are within the realm of 
possibility, as stated elsewhere in this dicussion, there are limits of practicality; consider all the 
information and real-time required for a person to perform those two actions; why expect a computer to do it 
faster??

Actually, the answer is that a computer should be able to get exposure to information at a >much< faster rate 
than people, but do consider the aspect of information coherancy; the information that people are exposed to to 
allow them to develop such skills is [generally] highly coherant; people don't sit in sensory depravation 
chambers reading newspapers all the time.  If they were to do so, I could virtually _guarantee_ you they would 
NOT be able to come up with a joke!  :)

A thinking neural net should be able to come up with a joke.....babys are totally dependant on their parents to 
provide for them...who has the patence to baby-sit a neural network?   You have reasonable assurance that a 
child is going to grow up to a thinking person; we have some millions of years of history to suggest that it 
will happen, but there is _no_ history to suggest that _any_ program, given _any_ amount of TLC will grow up to 
be a thinking application......so, given we believe our design to be able to think, we do believe our design has 
the capability to make up a joke, recognize a joke, etc....but there is training required!

BTW, after working on this for eight years, I can take that fressness statement as nothing but a compliment.

	>You may want to spend some time designing and getting specific 
	>measureable results with your project. Adding this dimension to the work 
	>you've already done may yield some useful perspectives. Since you are 
	>attracted to big challenges... how about a demonstration of something 
	>"simple" like speech recognition?

I am glad to >now< be able to say I have been working on the implementaion, and it is able to initialize a 
network.  I can now continue finishing making it iterate on a given set of input and/or training data.

the "simple" task of speech recognition has always been a particular interest of ours.  The PC hardware these 
days should be up to snuff to do that in real time.  I need to upgrade my personall equipment; a dual-P^ with a 
4 gig HD may suffice or at least be a good starting point.  Training for speech regognition is rather straight 
forward and well within the realm of what I can do as a personal effort (or, if I get my friend to work with me 
more, what WE could do).

Regards,
Joseph

