Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/5/84; site mnetor.UUCP
Path: utzoo!utcs!mnetor!sophie
From: sophie@mnetor.UUCP (Sophie Quigley)
Newsgroups: net.ai
Subject: Re: AI and Turing's Thesis
Message-ID: <839@mnetor.UUCP>
Date: Tue, 21-May-85 10:54:39 EDT
Article-I.D.: mnetor.839
Posted: Tue May 21 10:54:39 1985
Date-Received: Tue, 21-May-85 12:17:12 EDT
References: <113@nvuxf.UUCP>
Organization: Computer X (CANADA) Ltd., Toronto, Ontario, Canada
Lines: 40

> According to the Turing/Church-Kleene/Post Thesis (and derivatives thereof),
> "Any computation one might naturally regard as possible to carry out can
> be performed by some Turing machine having a suitable set of instructions"
> (All quotes in this posting are from "Machines, Languages, and Computation"
> by Denning, Dennis, and Qualitz), or in other words, anything which is 
> "computable" may be simulated by a Turing machine.
> 
> If we are attempting to "simulate" human intelligent activity on a 
> machine (activity including composing music), we must be assuming 
> that such activity is therefore "computable," if we accept that anything
> our machines do is a computable activity.
> 
> But Turing also stated that "The halting problem for Turing machines is
> unsolvable" where the halting problem is defined as "Given a Turing
> machine M and an initial tape T, does M halt when started on tape T?"
> I take all this to mean that one computer, given the inputs of
> another computer's hardware and software specifications and its input,
> cannot determine if the program will halt, or will go into "infinite
> loop."  In other words, one computer cannot "debug" another.
> 
> Human beings can "debug" programs, and are capable of determining if
> a program goes into "infinite loop."  If human beings can perform an
> activity that Turing machines cannot, they must not be computable.
> Therefore, human intellectual activity cannot be simulated on a machine.
> 
> There are three solutions that I see to this problem.  First, what the Thesis
> defines as "computable" is wrong.  Second, perhaps the machines we are
> developing today are more powerful than Turing machines, which begs the
> question of what is truly "computable."  Third, perhaps humans cannot
> always "debug" programs.  Douglas Lenat has stated that "AM" did things
> that he did not suspect.  Perhaps computable process Lenat was finding that
> it could not determine the "halting function" for Turing machine "AM."
> 
There is a fourth solution, the one you stated in the previous paragraph:
it is possible that human intellectual activity cannot be simulated on a
machine.  (I personally think that this is quite probable anyway, even
without worrying about Turing machines).
-- 
Sophie Quigley
{allegra|decvax|ihnp4|linus|watmath}!utzoo!mnetor!sophie