Path: utzoo!utgpu!jarvis.csri.toronto.edu!clyde.concordia.ca!uunet!cs.utexas.edu!usc!apple!amdahl!kp
From: kp@uts.amdahl.com (Ken Presting)
Newsgroups: comp.ai
Subject: Re: Why the Chinese Room doesn't convince
Summary: Some simple questions depend on semantics
Keywords: meaning, symbol, understanding
Message-ID: <70Qw02h57f9901@amdahl.uts.amdahl.com>
Date: 16 Jan 90 17:41:12 GMT
References: <18883@bcsaic.UUCP> <1863@osc.COM> <1623@castle.ed.ac.uk>
Reply-To: kp@amdahl.uts.amdahl.com (Ken Presting)
Organization: Amdahl Corporation, Sunnyvale CA
Lines: 28

In article <1623@castle.ed.ac.uk> arshad@lfcs.ed.ac.uk (Arshad Mahmood) writes:
>
>What would the chinese room say to being asked what was my last question ?
>
>For the room to be able to answer the question the person in the room
>must now be able to at least tell what this set of symbols means !, one can 
>quite correctly say that this can happen as an answer to a host of questions
>but I would contend that they are non-trivially similar and the semantics
>which has leaked out is fatal to the argument.

You have added to the category of Chinese Room questions which show that
no "pure abstract symbol system" could pass the Turing test.

Searle's argument depends on the assumption that symbols go in, get
syntactically manipulated, and new symbols go out.  Any CR which has a
rule book which specifies non-syntactic operations would IMHO not be a
purely syntactic device.  Now if no purely syntactic CR could pass the
Turing test, then Searle loses an essential premise in his argument.

Searle could perhaps reply that this objection just makes his burden of
proof easier.  If he wants only to show that pure syntactical processing
cannot cause thinking, he would be right.  But of course no advocate of
Strong AI would suggest that a program stripped of all transducers,
memory, and TOD clocks could pass the Turing test.

I've heard Searle speak on the subject of the Chinese Room, but objections
in this category were not raised.  Has anyone else heard him take up the
issue of implementation-grounded semantics?