Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!newshost.marcam.com!zip.eecs.umich.edu!newsxfer.itd.umich.edu!gatech!swrinde!cs.utexas.edu!rutgers!argos.montclair.edu!hubey
From: hubey@pegasus.montclair.edu (H. M. Hubey)
Subject: A More Stringent Test for Machine Intelligence
Message-ID: <hubey.786480314@pegasus.montclair.edu>
Keywords: Reverse the roles of humans and machines
Sender: root@argos.montclair.edu (Operator)
Organization: SCInet @ Montclair State
Date: Sat, 3 Dec 1994 18:45:14 GMT
Lines: 51



I got a thought last night about how it might be possible to allay
the reservations of anti-AI folks regarding the usefulness of the
Turing Test (TT).  

The TT essentially works by allowing humans to act as instruments
with which to measure intelligence.  Only the false alarms (i.e.
Type I and Type II errors of statistical testing) is used in
the test. Obviously if we had all the information on the populations
being tested --like the probability distributions, the means and
the variances, etc-- then we could use these  numbers to test if a 
given sample (batch) came from the human population or the machine
population. IN such a test, if it existed, we know --intuitively--
that the closer the means of the populations to one another the
more difficult it is for the test to determine from which population
the sample was obtained. In the limit (as both populations converged
i.e. identical distributions), then the test would be useless because
it would not be able to distinguish/discern the differences.

It is this part that the TT uses -- in a non-numeric sense --
i.e. we let the humans do what they want to tell us if the entity
being tested is human or machine.  If there are as many errors in
both directions, then we decide that the machine is indistinguishable
from humans.


It is this part we can change. Let us think of a variation of this 
test, say HT.  In the TT we use humans as instruments with which to measure
intelligence. In the HT we reverse the roles. We ask the machine to act
as the instrument to test the entity (on the other side of the wall) and
to tell us if it's human or machine.  If the performance of the
machine is as good (on the average) as humans on the same test, then
some of the criticisms of the test based on such things as, free will,
initiative, consciousness, awareness, etc, will be answered. After all,
in order to be able to actually ask questions, decide on what kinds
of questions should be asked next, etc, then machine would seem to be
displaying some kind of a capability of independent, goal-directed and
initiative-taking or creative endeavor.

I welcome comments.


PS. I'm not sure that it will really be a more stringent test of human
intelligence but it does seem to answer the criticisms of the TT a 
little better. 

--
						-- Mark---
....we must realize that the infinite in the sense of an infinite totality, 
where we still find it used in deductive methods, is an illusion. Hilbert,1925
