Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!oliveb!apple!voder!pyramid!prls!philabs!linus!mbunix!bwk
From: bwk@mbunix.mitre.org (Barry W. Kort)
Newsgroups: comp.ai
Subject: Re: Question on Chinese Room Argument
Summary: Discovering a missing component in the Chinese Room.
Keywords: Sentience, Knowledge Acquisition, New Ideas
Message-ID: <45445@linus.UUCP>
Date: 26 Feb 89 15:31:14 GMT
References: <45126@linus.UUCP> <5662@homxc.UUCP> <45199@linus.UUCP> <7219@polya.Stanford.EDU>
Sender: news@linus.UUCP
Reply-To: bwk@mbunix.mitre.org (Barry Kort)
Organization: Garden Golems, Inc., Norbert, WI
Lines: 47

In article <45199@linus.UUCP> I wrote:

 > The Chinese Room is like Helen Keller before her moment of
 > epiphany.  There is little point in manipulating symbols
 > mechanistically unless one can map the symbols to non-symbolic
 > sensory information from the external world.

In article <7219@polya.Stanford.EDU> geddis@polya.Stanford.EDU
(Donald F. Geddis) responds:

 > That might be true if the issue were learning.  In the case of
 > Searle's Chinese Room argument, however, we are already *assuming*
 > that the system is capable of communicating like a native speaker.
 > The system already acts as though it connected the symbols to their
 > non-symbolic referents.  Note how easy it was to know that Helen
 > Keller did *not* understand the connection: almost any simple
 > "conversation" gave it away.

Donald, I think we have uncovered an important issue hidden in the
Chinese Room debate.  When I have a conversation with another
intelligent being, I expect to exchange knowledge, such that we
both understand more than we did before the conversation.  That is,
I cannot conceive an intelligent entity which does not engage in
learning (knowledge acquisition).  When I add a symbol (such as
the word "colligate") to my personal lexicon, I also add its
referent.  Now when I sort through a jumbled collection of ideas,
trying to put the pieces together, I can associate that activity
with the word "colligation".  Can the Chinese Room do that?

 > Now it might be true that a computer system could not converse intelligently
 > without being embodied in the real world.  But the real question Searle
 > considered was:  How do you determine when a system is intelligent, when it
 > actually thinks?  The AI answer is "treat it as a black box and observe its
 > behavior (have conversations, in this case)".  Searle (mistakenly) disputes
 > this view, and wants us to look inside the system for some "causal powers".

I like this operational definition of intelligence.  I also believe
that if the candidate system were doing nothing more than formal
symbol manipulation, I could unmask it as easily as Feynman
unmasked the Brazilian physics students.  Formal symbol manipulation
is an important and useful tool for the cognitive computer, and any
intelligent entity is well-advised to acquire such capacity.  But
any intelligent system who thereupon stops learning is doomed to be
disparaged as lacking in a desirable quality:  the ability to
discover and report interesting new ideas.

--Barry Kort