Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!oravax!daryl
From: daryl@oravax.UUCP (Steven Daryl McCullough)
Newsgroups: comp.ai
Subject: Re: Chinese Room by Shannon and McCarthy from 1956
Message-ID: <1307@oravax.UUCP>
Date: 1 Feb 90 15:24:58 GMT
References: <2891@bingvaxu.cc.binghamton.edu>
Organization: Odyssey Research Associates, Ithaca NY
Lines: 62
Summary: Do we say "That isn't thinking" or "That isn't the way *we* think".

In article <2891@bingvaxu.cc.binghamton.edu>, cjoslyn@bingvaxu.cc.binghamton.edu (Cliff Joslyn) writes: (quoting Shannon and McCarthy)
> 
> "A disadvantage of the Turing definition of thinking is that it is
> possible, in principle, to design a machine with a complete set of
> arbitrarily chosen responses to all possible input stimuli.  Such a
> machine, in a sense, for any given input situation (including past
> history) merely looks up in a 'dictionary' the appropriate response. 
> With a suitable dictionary, such a machine would surely satisfy Turing's
> definition, but does not reflect our usual intuitive concept of
> thinking.  This suggests that a more fundamental definition must involve
> something relating to the manner in which the machine arrives at its
> responses -- something which corresponds to differentiating between a
> person who solves a problem by thinking it out and one who has
> previously memorized the answer".

I'm not certain that it is important to differentiate between these
two cases. The usual reasons that we worry when someone is using the
"wrong" method to solve problems are (1) there may come a time when he
is faced with a problem that is not in the set he memorized, and (2)
the correct method is important in its own right, since it teaches
general principles which will be useful in similar problems.

In both cases, the worry is that, although the problem-solver got the
right answer, there will eventually come a time when his
problem-solving performance will not be as good as someone who learned
the correct method. When it is clear that future performance is not
affected, noone cares (usually) whether answers are memorized or not.
Test yourself: if I ask you "What is 6 times 7?" do you figure out,
starting from the definition of multiplication, or do you recite a
memorized answer?

If a machine passes the Turing test, then by the definition of
passing, there is *no* performance difference between it and someone
who can *really* think. So why should we care *how* it does the
thinking?

There is also a very practical side to this question: a lookup table
for all possible input histories would be absolutely enormous! Consider
the assignment "Read this 10,000 word essay and write a report on it."
Assuming the essayist has a very small vocabulary of 1000 words, there
are still

         10,000
     1000

possible essays. A table of all possibilities would have many more
entries than there are atoms in our galaxy. For this reason, an actual
machine which could pass the Turing test would *have* to do something
more intelligent than a table lookup. 

O------------------------------------------------------------------------->
> | Cliff Joslyn, Cybernetician at Large,cjoslyn@bingvaxu.cc.binghamton.edu 
> | Systems Science, SUNY Binghamton, Box 1070, Binghamton NY 13901, USA 
> V All the world is biscuit shaped. . .

. . . Sure. And I've got one, two, three, four, five
senses working all the time.

Daryl McCullough, Odyssey Research Associates, Ithaca, NY
oravax.uucp!daryl@cu-arpa.cs.cornell.edu