From newshub.ccs.yorku.ca!ists!torn.onet.on.ca!utgpu!news-server.csri.toronto.edu!rpi!gatech!mcnc!aurs01!throop Tue Jun 23 13:20:47 EDT 1992
Article 6267 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!torn.onet.on.ca!utgpu!news-server.csri.toronto.edu!rpi!gatech!mcnc!aurs01!throop
>From: throop@aurs01.UUCP (Wayne Throop)
Newsgroups: comp.ai.philosophy
Subject: roles in the turing test
Message-ID: <60829@aurs01.UUCP>
Date: 16 Jun 92 18:26:51 GMT
References: <1992Jun12.194443.37383@spss.com>
Sender: news@aurs01.UUCP
Lines: 47

> markrose@spss.com (Mark Rosenfelder)
>> throop@aurs01.UUCP (Wayne Throop)
>>I'd probably answer something like "I regard that question as cheating. 
> Interesting idea-- that the *judge* can cheat on the Turing Test.  If
> we take it to its logical conclusion, we'll see that the Turing Test
> cannot be used to define intelligence.

It seems to me that if the judge peeks behind the partition that
separates the judge from the testees/controls, and the judge bases a
decision on what is seen there, then that judge has quite clearly
cheated.  Not so?  What makes this "interesting"?

( I see basically four roles in the turing test: judge, testee,
  control, and moderator.  The moderator sets up the partition, verifies
  that they do indeed hide the testees and controls adequately, fetches
  ice water for the controls and machine oil for the testees, and so on.
  There might also need to be other roles if the test is to be adequately
  double-blind. )

And while it isn't clear to me that the TT *ought* to be used to
define intelligence, I'm at a loss to see how "judge cheating" of this
form leads to the conclusion that it *can't* be used so (which
presumably means that the test results are ill-defined in some sense
or other).

( Note that I now agree that such an answer would be self-defeating
  for a testee.  It isn't yet clear to me one way or another whether 
  a moderator "should" censor the question  (that is, whether the
  question should be regarded as an attempt to cheat). )

> Presumably the idea here is that the judge's question is *irrelevant* to
> the purpose of the test.

No, not quite.  I saw some reason to suppose that questions like that
are a violation of the basic form of the test, specifically that the
physical forms of the testees and the controls are concealed.  So, not
so much irrelevant to the test, but explicitly ruled illegal by the
test.  Or another way: I'm not going along lines of "everything not
explicitly allowed is illegal", but rather that I imagined this would be
explicitly disallowed in a fully-spelled-out set of TT rules.

So, do questions of this form fall under the "imitating human
conversation" or under "attempting to peek behind the curtain"?
(It seems most people disagree with me, and just naturally see it
as part of the former, while I saw it as part of the latter.)

Wayne Throop       ...!mcnc!aurgate!throop


