From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!usc!wupost!darwin.sura.net!gatech!mcnc!ecsgate!lrc.edu!lehman_ds Tue Jan 28 12:15:25 EST 1992
Article 2986 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!usc!wupost!darwin.sura.net!gatech!mcnc!ecsgate!lrc.edu!lehman_ds
>From: lehman_ds@lrc.edu
Newsgroups: comp.ai.philosophy
Subject: Re: Table-lookup Chinese speaker
Message-ID: <1992Jan21.160425.129@lrc.edu>
Date: 21 Jan 92 21:04:25 GMT
References: <1992Jan18.134742.4155@oracorp.com> <1992Jan20.182835.5307@spss.com>
Organization: Lenoir-Rhyne College, Hickory, NC
Lines: 44

In article <1992Jan20.182835.5307@spss.com>, markrose@spss.com (Mark Rosenfelder) writes:
> Your replies don't address my point, which is that a table-lookup Chinese
> speaker is IN PRINCIPLE impossible.
> 
> The mere fact of possessing a database of successful conversations does NOT
> imply that the machine can itself pass the Turing test.  The basic problem
> is that the machine's responses can be constrained to lie in the set of
> successful conversations, but the human's cannot be.
> 
> Let's call the set of possible (up to hundred-year) conversations S.
> Within S we enumerate as T the set of conversations we judge to have passed 
> Turing Test.  It does not matter what criteria we use-- we can be as cautious
> or as generous as we like.
> 
> Now picture a conversation C which fails the Turing test (that is, it's in
> S but not in T).  We can represent C as s(0), s(1), s(2), ..., where these
> are particular statements.
> 
> C is not in T, but initial sequences of C must exist in T.  For instance,
> there must be a conversation which begins with statement s(0).  (If there
> isn't, the machine immediately fails the Turing test, since you could stymie
> it by beginning the conversation with s(0).)  Similarly, there is probably
> a successful conversation which begins with s(0), s(1).  
> 
> There is some number m which is the highest number such that a conversation
> beginning s(0), s(1), ... s(m) exists in T.  (m must be less than the total
> length of the conversation, because we said that C was not in T.)
> 
> We can therefore have a conversation with the machine which starts with
> s(0), s(1), ... all the way up to s(m).  The computer can utter s(m)
> because it has in its table (that is in T) at least one conversation
> (needless to say one besides C) which begins with this sequence.
> 
> However, we now utter s(m+1).  The machine now has no valid response to make.
> There is no conversation in T which begins s(0), ... s(m), s(m+1), since m
> is the *largest* number such that a conversation s(0), ..., s(m), ...
> exists in T.  Since the machine has no response to make, it fails the
> Turing test.
  This is where table look-up fails, but not where AI fails.  If a machine
had a set of rules to produce a conversation, then the machine would be
able to adapt to the s(m+1) and continue on into infinity, would it not?
  By breaking a sequence down into a function we can represent any s(n).
    Drew Lehman
    Lehman_ds@mike.lrc.edu


