From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!garrot.DMI.USherb.CA!uxa.ecn.bgu.edu!mp.cs.niu.edu!linac!uchinews!spssig.spss.com!markrose Thu Feb 20 15:21:30 EST 1992
Article 3806 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!garrot.DMI.USherb.CA!uxa.ecn.bgu.edu!mp.cs.niu.edu!linac!uchinews!spssig.spss.com!markrose
>From: markrose@spss.com (Mark Rosenfelder)
Newsgroups: comp.ai.philosophy
Subject: Re: Look-up tables
Message-ID: <1992Feb17.193928.65482@spss.com>
Date: 17 Feb 92 19:39:28 GMT
References: <299C3CC6.15601@orion.oac.uci.edu> <1992Feb15.114130.8837@husc3.harvard.edu> <299DAD7F.14042@orion.oac.uci.edu>
Organization: SPSS Inc.
Lines: 53
Nntp-Posting-Host: spssrs7.spss.com

In article <299DAD7F.14042@orion.oac.uci.edu> schoi@teri.bio.uci.edu (Sam "Lord Byron" Choi) writes:
>Furthermore, why are you assuming that all questions or parts of a
>question are going to make any sense at all?  The purpose of the tester
>in the turing test is to try and figure out whether the interlocutor is
>a human being or a computer.  Human beings are perfectly capable of
>dealing with ambiguous, stupid or entirely non-sensical questions.
>
>If someone came up to me and said, "who, the man, five times,
>running what that and" I would say, "what the fuck are you talking
>about?"

Right; so that's a reasonable response to put in the table.  The
table contains at least one reasonable response for every possible input.

>Any cludge which would allow the computer to respond to something
>that is not on its look-up table with "I don't understand what you are
>talking about" is no longer a program which relies solely on a look-up
>table for its responses.

*Everything* is in the computer's lookup table.  This seems to be what
you're not scoping on.

>How many numbers are there between 1 and 2?
>
>Can a finite look-up table possibly hope to list all of their possible uses
>in any possible conversation?
>
>I could ask "how many times does 1.2343987453647563 go into 2.1?"
>A human respondant might say "who knows . . . well, more than 1 and
>not quite two."
>
>Is every single version of this question going to be included in your
>look-up table?
>
>I think not.

Here's how we construct the table.  We consider every possible opening
question from the tester.  Yes, *every possible statement*, sensible or not.
There is a finite number of statements which can be made under our time
limitation (100 years), so this is a finite task.  For each statement
we come up with (at least) one reasonable reply, and enter it into the
table.

Then we consider *every possible second statement* the human tester can
come up with, and enter into the table a reasonable response to it (fitting
into the conversation so far).  And so on.

You seem to be having trouble with the concept that the set of possible
statements is finite.  But it is; a finite number of sentences can be
arranged into texts of finite length in only a finite number of ways.

(Naturally, the total number of texts is so large that the lookup table
is completely impossible in practice.)  


