Re: Turing Test

From: HARNAD Stevan (harnad@cogsci.soton.ac.uk)
Date: Wed Jun 05 1996 - 18:04:52 BST


> From: "Dearns, Rachel" <rld195@soton.ac.uk>
> Date: Sat, 25 May 1996 14:33:24 +0100 (BST)
>
> The Turing Test is basically a critical test of whether or not a
> computer has equal intelligence to a human.

No, it's a test of whether it has "intelligence" (= a mind) AT ALL.

> The test proposes the
> situation of having a person in one room and a computer in another,
> both could be communicated with in a similar way to communicating with
> a penpal. The communication could be about anything, for any length of
> time, and the test states that it would be impossible to tell which of
> the participants was the human, and which was the machine.

The test does not STATE this, it tests whether or not it is true that
you cannot tell them apart.

> This being
> the case, the conclusion of this test is that if you can't tell the
> difference between the two participants after the communication then
> you cannot say that a machine does not have a mind. Turing, who
> suggested the test, stated that a person communicating with the
> computer would never even suspect that the other two participants were
> not both people

Hi said that IF you would never suspect otherwise (for a lifetime)...

> - and if you subsequently found out it would be
> arbitrary to deny that the candidate was intelligent simply because you
> had been informed that it was a computer. It is important to note that
> the computer is kept out of site in the Turing test so that the
> judgement made would not be biased by what the machine looked like
> (ie. that it did not look like a human.) However no Turing test can
> guarantee that a body has a mind. Nothing of the explanation of its
> successful use requires that the model involved needs to have a mind at
> all. Also, this original 'penpal' version of the Turing test only tests
> linguistic capacity.

The penpal point is valid: What about our capacity to interact with real
objects in the world? This calls for a robot version of the Turing Test.
Also, the reason the Turing Test can't guarantee that the penpal has a
mind is that NOTHING can guarantee this -- because of the other-minds
problem.

> In summary the Turing test states in some way that if a machine can do
> everything a body with a mind can do, such that the person and the
> machine cannot be told apart; then we can have no basis for ever
> doubting that the machine has a mind.

We have no better or wrose basis for doubting it than we have for
doubting it in one another.

For an A, relate to the other-minds problem, reverse engineering, and
causality.



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:45 GMT