Re: Turing Test question

From: HARNAD Stevan (harnad@coglit.ecs.soton.ac.uk)
Date: Sat Mar 31 2001 - 11:46:48 BST


On Fri, 30 Mar 2001, Hugh Gene Loebner wrote:

> Hello Dr. Harnad,
>
> I believe that Alan Turing intended his test to be more general than is
> recognized.* That was one reason I included the requirement that the
> winning program must deal with audio/visual input.

Dear Hugh,

Nice to hear from you. First, may I have permission to include your
message and my reply in the Skywriting Archive I maintain for my
students on this (and related) topics at:

http://www.cogsci.soton.ac.uk/~harnad/Hypermail/Foundations.Cognitive.Science2001/

Second, A/V input is certainly a good idea, but probably motor output,
and indeed real-time sensorimotor interactive capacity with the objects
and events in the world is necessary to make the TT rigorous and
exhaustive. (I suspect even more: that this robotic capacity is needed,
even if it is not directly tested, even to pass the original "pen-pal",
symbols-only version, which is the one most people, including the
Loebner Prize Committee, consider to be the "official" TT.)

> I argued with the Loebner Prize Committee (while there was a LP
> Committee) to include robotics as a criterion, but got nowhere. Later,
> after the LP Committee resigned en masse, I set forth my requirement
> for A/V, which precipitated a fire storm of protest from the usenet AI
> community. I just didn't have the strength to go for robotics.

They protested, because a one-night, one-shot pen-pal programme is a
lot easier to write than a viable robot is to design!

> If there are any flaws in the LP contest beyond whatever intrinsic
> flaws may lie in the TT test, please advise, since I have tried to make
> the LP contest as "pure" a TT test as possible (at least a verbal TT).

There is a flaw in the TT at least as profound as restricting it to the
pen-pal version, and that is restricting it to a one-shot, one-night
affair (like the LP)! The TT is meant to confer REAL performance
capacity, equivalent to that of a human with real intelligence, and
indistinguishable from that of a human with real intelligence -- but
not for 2 minutes, or 5 minutes, or one evening, or for the duration
of an LP contest! Turing (1950) was being not only arbitrary but
downright silly when he put it thus:

    "I believe that in about fifty years' time it will be possible, to
    programme computers, with a storage capacity of about 109, to make
    them play the imitation game so well that an average interrogator
    will not have more than 70 per cent chance of making the right
    identification after five minutes of questioning."

    Turing, A. M. (1950) Computing Machinery and Intelligence. Mind
    49:433-460.
    http://cogprints.soton.ac.uk/documents/disk0/00/00/04/99/index.html

Not only is the TT not about "5-minute" capacity (but rather about
real, total, life-long capacity): it is also not about "fooling" 70% of
the people 70% of the time! It is about generating the real capacity,
totally equivalent to and indistinguishable from our own, to anyone,
for as long as it takes! Otherwise it is not just arbitrary, but
trivial. (There are plenty of things that fool 70% of the people 70% or
even 100% of the time: So what?)

See:

Harnad, S. (2000) Minds, Machines, and Turing: The Indistinguishability
of Indistinguishables. Journal of Logic, Language, and Information 9(4):
425-445. (special issue on "Alan Turing and Artificial Intelligence")
http://www.cogsci.soton.ac.uk/~harnad/Papers/Harnad/harnad00.turing.html

> (*) As you must know, Turing discussed storage requirements of the test
> when played with a blind person, and he suggested buying sense organs
> for heuristic programming of the machine. I think these are sufficient
> evidence of his interest in sensory input.
>
> Regards,
>
> Hugh Loebner

But the wrong kind of evidence, if all he wanted was to include blind
interrogators among his 70%. The fact is, sensorimotor capacity is not
only an essential part of human intelligence capacity, but chances are
that the symbolic part of human intelligence capacity is grounded in the
sensorimotor part!

Best wishes,

Stevan
--------------------------------------------------------------------
Stevan Harnad
Professor of Cognitive Science
Department of Electronics and phone: +44 23-80 592-582
             Computer Science fax: +44 23-80 592-865
University of Southampton http://www.cogsci.soton.ac.uk/~harnad/
Highfield, Southampton http://www.princeton.edu/~harnad/
SO17 1BJ UNITED KINGDOM



This archive was generated by hypermail 2.1.4 : Tue Sep 24 2002 - 18:37:30 BST