Re: the Turing test

From: Hal Finney (hal@rain.org)
Date: Fri Oct 18 1996 - 13:12:39 MDT


For information about the current state of the art in Turing Test
passing programs, see the Loebner Prize home page, at <URL:
http://acm.org/~loebner/loebner-prize.htmlx >. There you will find
links to recent winners, including the 1996 winner, a program from
Australia. There are actually two programs there, one called HeX and
the other called (I am somewhat embarrassed to say) HAL.

These programs are all smoke and mirrors, using conversational
trickery to try to distract and misdirect the judge for long enough
that he doesn't notice the utter lack of understanding which the
program really possesses. It's kind of like watching David Letterman
fly. You might not be able to distinguish him from someone really
flying given the constraints of limited time and viewing angles, but
that doesn't mean there's no difference.

When you look at the larger body of transcripts you begin to see the
repetitions, the errors, the flashy but canned statements, and you
realize how little is actually there.

The author readily admits this, and on his web page you can find his
article called "How to Pass the Turing Test by Cheating."

Lyle's game is amusing but as in many such cases the facts are ultimately
more interesting. The real question is whether the Turing test is valid,
and in particular just how much interaction is necessary before we can
know that the program is showing real understanding.

Hal



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:35:47 MST