From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!cs.utexas.edu!wupost!uwm.edu!ogicse!qiclab!nosun!hilbert!max Tue Nov 19 11:09:17 EST 1991
Article 1227 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!cs.utexas.edu!wupost!uwm.edu!ogicse!qiclab!nosun!hilbert!max
>From: max@hilbert.cyprs.rain.com (Max Webb)
Newsgroups: comp.ai.philosophy
Subject: Re: Animal Intelligence vs Human Intelligence
Message-ID: <1991Nov7.010130.28204@hilbert.cyprs.rain.com>
Date: 7 Nov 91 01:01:30 GMT
References: <37443@shamash.cdc.com> <1991Oct31.235402.12739@hilbert.cyprs.rain.com> <37658@shamash.cdc.com>
Organization: Cypress Semiconductor Northwest, Beaverton Oregon
Lines: 81

In article <37658@shamash.cdc.com> map@svl.cdc.com writes:
>In <1991Oct31.235402.12739@hilbert.cyprs.rain.com> max@hilbert.cyprs.rain.com (Max Webb) writes:
>
>>In article <37443@shamash.cdc.com> map@svl.cdc.com writes:
>>Further, I have seen video tape where Washoe, confronted with her
>>doo-doo on the carpet, looks around and names various people as
>>the perpetrators...
>>The only explanation that makes sense to me is that
>>	1) she knew she would be punished for that, if someone
>>	   else knew.
>>	2) she knew if she could convince someone else that it wasn't her,
>>	   she wouldn't be punished.
>What is there in this   
>incident that requires conceptualization by Washoe?
>Everything element of it is perceptual - the people, Washoe, the doo-doo,
>the memory of punishment for past transgressions.

That doesn't wash. The utterances only avoid punishment in a narrow
set of cases. Unless the chimp can quickly learn to lie in only (or mostly) 
these cases, the punishment/reward incentives never even get a chance
to work, because the net reward is negative.

	1) (Requirement)
	   For "Mark did it" to succeed in avoiding punishment,
	   it is necessary to determine whether the act "it" was good or bad.
	   Otherwise, it may be a reward that was avoided.

	   (behavior description)
	   This demonstrates that something in Washoe 
	   maps the pronoun to the correct act, and then determines
	   whether saying 'I did it' will result in reward or punishment.
	   (since lying appeared full blown, with the good act-bad act
	   distinction already in place. There was simply no chance for
	   Washoe to learn the map:
		Just_did_bad_thing && utter(I_did_it) => punishment
	   if you treat the string as a single opaque unit to washoe).

	   (conclusion)
	   This use of the pronoun is a fairly sophisticated
	   linguistic behavior, whose reward/punishment is context
	   dependent in a complex way. No simple S-R link can
	   explain it, any more than it can explain the learning of
	   a potentially infinite set of utterances in all natural
	   languages.

	2) (Requirement)
	   Washoe must not have been witnessed. If the act of lying cannot
	   convince it will not be rewarded, but _punished_. (as it was
	   in this case).

	   (Behavior)
	   For the behavior to be successful more than half the time,
	   something in Washoe must permit the successful distinction
	   between plausible lies and implausible lies more than half
	   the time. Doing this successfully requires some representation
	   of the beliefs/knowledge of others and how this motivates them.

	   (conclusion)
	   I don't see how you can do that without some representation
	   of very high level concepts - the difference between
	   beliefs and reality, and the way beliefs motivate behavior.

>  The pleasure-pain
>mechanism is basic to life, and I don't see any reason to look elsewhere for
>an explanation for this incident.

It appears to me that you ignore the structure and context dependence
of the behavior, which could only be implemented with _some_ representation
of the belief systems of others. You point to the reward/punishment mechanism
as if that was all that was required.

Look a little closer. Lying is very sophisticated behavior.
There is no clear cut 'Response'. No clear cut 'Stimulus'.
Your argument is exactly analogous to the analyses of
HUMAN language advanced by behaviorists in the past, and
fails (IMHO) for the same reasons.

>--
>Mark A. Peters                              ****** ======================
>Control Data Corporation                    ****** == "What a save!!!" ==
>Internet: map@svl.cdc.com                   ****** == "What an idea!!" ==


