Turing, Giving the Devil His Due

[From Rick Marken (941118.1000)]

Bruce Nevin (Wed 931116 09:22:11 EST) --

Has the CSG-L server kicked the cosmic bit bucket?

It's hard to tell whether it's the server or the servees.

I thought the relationships between the Turing Test and the Test might be
amusing.

I think this could be a most illuminating topic. The continued homage paid to
the Turing Test by cognitive scientists proves that cognitive psychology is
pure behaviorism: superficial appearances are the basis for recognizing
"intelligent behavior". If it walks like a duck, quacks like a duck,
and shits like a duck, then (according to the Turing Test) it's a duck. The
Turing Test is based on the idea that the best model of a living system is
the one that mimics the most superficial appearances of the system's behavior
and/or does this mimicing in the most convincing way. The Test (in PCT) adds
one extra criterion to the Turing Test: it not only has to walk like a duck,
quack like a duck, and shit like a duck, it also (and most importantly) has
to WANT what a duck wants. The Test is the way you find out what living
systems WANT -- that is, what perceptual variables they control.

Bruce Abbott (941117.1115 EST) --

Apparently I have been playing Devil's advocate all too well--people are
starting to suspect that I am in fact the Devil.

Tom Bourbon (941117.1617) and Bill Powers (941118.0830 MST) already had some
nice comments about this. I just want to say that I am a big fan of Devil's
advocacy; it's good to have honest and forceful representations of opposing
points of view. But there is a problem with Devilish advocacy with respect
to PCT because PCT nearly always opposes the Angels that the advocate is
secretly defending as well. This happens because both Devilish and Angelic
behavioral theories tend to be based on the same foundation -- the cause-
effect model of behavior. For example, we have had other Devil's advocates on
this net advocating (devilishly, supposedly) stimulus-response models of
behavior. When they finally took off their red pajamas and pointed tail it
turned out that their angelic model was based on the idea that input
guides output -- the cause-effect model, with wings. You are devilishly
advocating reinforcement theory (a cause effect model) and angelically
embracing control theory as a model of selection by consequences (a cause-
effect phenomenon). That's why your Devil's advocacy doesn't always click
with me.

Perhaps the best way to start trying to clear this up is to start dealing
with the different predictions of reinforcement and control theories. Then we
don't have to worry about whether you are playing Devil's advocate or not; we
can just see the real (and dramatic) differences in observed behavior
predicted by these two models -- and eventually see which predictions match
the observed behavior.

Best

Rick

[From Oded Maler (941121)]

* [Rick Marken (941118.1000)]

ยทยทยท

*
* Bruce Nevin (Wed 931116 09:22:11 EST) --
*
* >Has the CSG-L server kicked the cosmic bit bucket?
*
* It's hard to tell whether it's the server or the servees.
*
* >I thought the relationships between the Turing Test and the Test might be
* >amusing.
*
* I think this could be a most illuminating topic. The continued homage paid to
* the Turing Test by cognitive scientists proves that cognitive psychology is
* pure behaviorism: superficial appearances are the basis for recognizing
* "intelligent behavior". If it walks like a duck, quacks like a duck,
* and shits like a duck, then (according to the Turing Test) it's a duck. The
* Turing Test is based on the idea that the best model of a living system is
* the one that mimics the most superficial appearances of the system's behavior
* and/or does this mimicing in the most convincing way.

I think you've got it wrong. The Turing test is just a humble way for
escaping the inexistence of an objective definition of what "intelligence"
is. In the absence of such a definition Turing suggested the following
challenge for creators of mimicking machines: make your machines
indistinguishable from humans by human observers. The question of whether
passing the TT implies intelligence/mind/etc. is another question (Searle's
chinese room etc.).

* The Test (in PCT) adds
* one extra criterion to the Turing Test: it not only has to walk like a duck,
* quack like a duck, and shit like a duck, it also (and most importantly) has
* to WANT what a duck wants. The Test is the way you find out what living
* systems WANT -- that is, what perceptual variables they control.

Ahem. There is nothing that prevents the interrogator in such a test
from using questions that intend to discover the "goal" of the human
(or program), and to base his response (is it a human or a machine) on
the answer to such questions. What you might be claiming is that if
the referee in such a Turing test was an omniscient PCTer (after
everything about higher levels has been resolved..) then he would
classify as "intelligent" and human-like only those machines which
were programmed according to PCT principles.

Bref, the TT and the Test treat two separate questions ( "does the subject
on the other side of the line dialog like an (*arbitray*) human?" and
"what perceptual variables does this *specific* subject control?")
These question are orthogonal, and I think that by now, there is enough
accummulated evidence on the net that the world does not divide solely
according to PCT non-PCT lines.

Regards,

--Oded

p.s.

Btw, I really symathize with your feelings about the recent demonstration
of vox populus in your, otherwise great, country.

--

Oded Maler, VERIMAG, Miniparc ZIRST, 38330 Montbonnot, France
Phone: 76909635 Fax: 76413620 e-mail: Oded.Maler@imag.fr

[From Jan Talmon (941122)

Rick Marken (941118.1000) writes:

Bruce Nevin (Wed 931116 09:22:11 EST) --

Has the CSG-L server kicked the cosmic bit bucket?

It's hard to tell whether it's the server or the servees.

I thought the relationships between the Turing Test and the Test might be
amusing.

I think this could be a most illuminating topic. The continued homage paid to
the Turing Test by cognitive scientists proves that cognitive psychology is
pure behaviorism: superficial appearances are the basis for recognizing
"intelligent behavior". If it walks like a duck, quacks like a duck,
and shits like a duck, then (according to the Turing Test) it's a duck. The
Turing Test is based on the idea that the best model of a living system is
the one that mimics the most superficial appearances of the system's behavior
and/or does this mimicing in the most convincing way. The Test (in PCT) adds
one extra criterion to the Turing Test: it not only has to walk like a duck,
quack like a duck, and shit like a duck, it also (and most importantly) has
to WANT what a duck wants. The Test is the way you find out what living
systems WANT -- that is, what perceptual variables they control.

....

Best

Rick

I think Rick is wrong here.
What Turing had in mind when he proposed the imitation game was to have a test
to decide whether computers can think. He never intended to consider the
mechnical behaviour.
As a matter of fact, the Turing test is a TEST in the PCT sense. In the Turing
test (not in the current contest !!) it is allowed to introduce disturbances
during the conversations. When the responses that appear on the terminal
cannot be discerned from responses given by a human, it is said that the
machine can think like a human. Although it sounds like S-R, there is no
objection to have the experimenter executing a PCT TEST.
It is likely that only a system implementing the higher levels of the HPCT
model can pass the Turing test.

It has been Stevan Harnad who has argued that the Turing test is insufficient
to decide whether machines can think. He argues that a system that processes
only symbols can never pass as a model for the human cognitive powers. Such a
system would need "connections" with the physical world to relate the symbols
to what is happening in the physical world: Also the lower levels of HPCT are
needed.
An object that walks, quacks and shits like a duck is only said to to be a
good model when it's behaviour in the real physical world - including variable
disturbances - is indiscernable for the behaviour of a duck.

It is (at least for the PCT believers) the S-R models that are used in the
existing systems that make them fail the Turing test. So Rick, why not quickly
implement a PCT model (it doesn't have to speak, it only has to control at
it's lowest levels the sequence of characters a dumbo in the physical world is
tying on the keyboard by spitting out characters on a screen). When your model
is OK, the first price of $100.000 would be a good start for the RIP
(Research/Rick's* Institute for Perceptual control).
* delete what is not applicable.

Jan Talmon
Dept. Medical Informatics
University of Limburg
Maastricht
The Netherlands