From newshub.ccs.yorku.ca!torn!cs.utexas.edu!sdd.hp.com!caen!umeecs!dip.eecs.umich.edu!marky Wed Aug 12 16:52:38 EDT 1992
Article 6581 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!torn!cs.utexas.edu!sdd.hp.com!caen!umeecs!dip.eecs.umich.edu!marky
>From: marky@dip.eecs.umich.edu (Mark Anthony Young)
Subject: Turing Test Myths
Message-ID: <1992Aug8.215358.2880@zip.eecs.umich.edu>
Sender: news@zip.eecs.umich.edu (Mr. News)
Organization: University of Michigan EECS Dept., Ann Arbor
References: <1992Jul16.093057.8880@techbook.com> <BILL.92Jul16201712@ca3.nsma.arizona.edu>
Date: Sat, 8 Aug 1992 21:53:58 GMT
Lines: 54


%r bfish@sequent.com (Brett Fishburne)
%q My argument is that tests which argue they define "intelligence" 
%q (such as the Turing Test) don't do an adequate job.

Turing never claimed that his test gave a definition of intelligence.  It is
merely an operational test (hence the name).  Furthermore, it is clear to me
that he meant it only to be a _sufficient_ test:  that is, if you pass, you're
intelligent; if you fail, we can't conclude anything.

(As evidence for my position, I offer the following quote:

    "May not machines carry out something which ought to be
     described as thinking, but is very different from what a
     man [sic] does?  This objection is a very strong one,
     but at least we can say that if, nevertheless, a machine
     can be constructed to play the immitation game 
     satisfactorally, we need not be troubled by this
     objection.")

%r szabo@techbook.com (Nick Szabo)
%q I've never liked the Turing test.  "Eliza" passed the Turing test
%q with some people, and it was just a babbling fake Rogerian therapist.

Eliza never passed the Turing Test.  Never, never, never, never, never.  Not
even once.  All we have is some anecdotes saying that some people thought they
were talking to a person when they were really talking to Eliza.  That hardly
counts as science!

Furthermore, it's meaningless to say that something "passed the Turing test
with some people".  The Turing test is a statistical test.  Thus the result is
a "confidence level" that the system has indeed passed the test.  

Say you actually ran the Turing Test, got five hundred subjects (including at
least one hundred each of men and women), ran the two versions of the
immitation game (one hundred instances of each).  Say that the men got
selected as women 37 times, and the machine as a human 44 times.  You might
say that the machine had passed the Turing Test, and many people would agree
with you.  But you would not be able to get your results published in a
respectable scientific journal UNTIL YOU HAD REPLICATED THE RESULT.

If the machine never came close again, staying down around 10 to 15 selections
thereafter, that would throw doubt on your original finding.  It wouldn't be
long before any statement of the form "X passed the Turing Test" would be met
with "That result was never replicated."  Eventually, people would stop saying
"X passed the Turing Test", and would instead say "They thought X passed the
Turing Test."  It would become a footnote in the history of computing.  

(I say this last bit even tho' I know it's not true.  After all, there are
still people who say that Eliza passed the Turing Test.  It's clear that
people will always talk rot.)

...mark young



