From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!ames!olivea!uunet!tarpit!cs.ucf.edu!news Thu Apr 30 15:22:25 EDT 1992
Article 5230 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!ames!olivea!uunet!tarpit!cs.ucf.edu!news
>From: clarke@next1.acme.ucf.edu (Thomas Clarke)
Newsgroups: comp.ai.philosophy
Subject: Re: SHRDLU's mind
Message-ID: <1992Apr22.154011.6453@cs.ucf.edu>
Date: 22 Apr 92 15:40:11 GMT
References: <1992Apr17.013517.15215@psych.toronto.edu>
Sender: news@cs.ucf.edu (News system)
Organization: University of Central Florida
Lines: 43

In article <1992Apr17.013517.15215@psych.toronto.edu> christo@psych.toronto.edu  
(Christopher Green) writes:
> In article <1992Apr13.151731.22702@oracorp.com> daryl@oracorp.com (Daryl  
McCullough) writes:
> >christo@psych.toronto.edu (Christopher D. Green) writes:
> >
> >> Surely whether SHRDLU, thermostats, and rocks have minds is an
> >> empirical question.
> >
> >Surely not. It is a philosophical question, which is the reason that
> >we have this newsgroup. It isn't an empirical question because it
> >doesn't have any observable consequences.
> >
> Surely so. To be empirical, a question must be a matter of fact, not just
> of definition. 

This is wonderful.  A concise statement of the argument in plain English.

Might I suggest the Turing test as a not-so original resolution. If a device  
passes the Turing test it has a mind.  This is an observable consequence, a  
matter of fact,  which can be determined empirically.

The difficulty of course lies with the empirical determination, especially at  
the margin.  Helen Keller and Stephen Hawking surely have minds, but the tester  
has to exert some effort to administer the test.

The technique used by Putnam seems to imply that even a rock, (hence also a  
thermostat) has a mind.  But his argument requires a vastly complex state  
transformer to administer the test.  This clearly is cheating.  The rock is  
redundant and the intelligence resides in the state transformer.

SHRDLU is somewhere in the middle.  It is not inconcievable that a procedure  
exists, say executable on a 10 MIPS workstation in real time, that can  
translate Turing test questions/answers to and from block world configurations  
and SHRDLU questions in such a way that SHRDLU passed the test.  Given this, it  
would follow that SHRDLU has a mind since the workstation translator does not  
add much to system.  If a Connection machine were required for the translation,  
the question of cheating arises and one would be led to question the locus of  
the intelligence.

Precisely defining what it means to cheat on a Turing test is a question for  
another day.  This problem is likely just the original argument in another  
guise.


