Newsgroups: comp.ai.nat-lang,comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!newsfeed.internetmci.com!news.dacom.co.kr!news.netins.net!duke!scottm
From: scottm@uis.com (Scott Miller)
Subject: Re: Losing the Loebner Competition Forced me to Re-evaluate my Humanity
Message-ID: <DKxI28.3xt@uis.com>
Date: Tue, 9 Jan 1996 19:27:43 GMT
References: <4cbrd5$odf@crc-news.doc.ca> <4ct9j5$kue@ralph.vnet.net>
Organization: Unix Integration Services
Lines: 26
Xref: glinda.oz.cs.cmu.edu comp.ai.nat-lang:4376 comp.ai:35750


These are some good points.  My feeling is not to worry too much about
whether these changes are "in the spirit of the Turing test".  After
all, it's expected that the contestants (the AI programs) would evolve
or improove over time; why shouldn't the methods to evaluate the 
contestants evolve (and hopefully improove) too?

				Scott Miller

tadpole@vnet.net writes:
>Very interesting.  It seems to me that one problem is that judges more or less
>"expect" to interact with computers.  That is, they expect some respondents to
>be computers, so they are forced to resort to "abnormal" questions, questions
>that people would not ask each other upon the first meeting.
>
>I wonder if it is possible to hold a Turing test where the judges do not know
>that they are judges until after the interaction.  After the interaction, 
>the judges could be told of the purpose, and have time to review transcripts.
>This spin might get rid of some of the oddball questions, and put the 
>judging back to a level field.  One must ask, however, whether such a 
>competition (which would be much more difficult to administer) is in the 
>spirit of the Turing test.

>Tom Gordon
>IBM Software Solutions, VisualAge and IBM Smalltalk

