Artifact Busters IV

[From Bruce Abbott (941127.1900)]

Rick Marken (941126.1000)

Bill Powers (941123.1415 MST)--

I don't know what you [Bruce] will make of this, but in fact the logical
path you describe just above is NOT the effective path-- it works the wrong
way, with a probability of 0.37.

By a chance property of the situation, however, this turns out to be the
least probable path, so it is most likely that this probability will be
decreased in the long run via a different path.

This is a typically subtle and insightful way of saying what I've been
trying to say all along. Bruce's "law of effect" model contains logic that
"strengthens" the "wrong" response (tumbling when going up the gradient)

So what? This is what the environment sometimes "rewards." An equivalent
statement in PCT is that the reason the control-model (non-learning) e. coli
tends to climb the nutrient gradient is because of the "artifact" that
tumbling during unfavorable nutrient rates tends to select more favorable
conditions, and that not tumbling during favorable nutrient rates tends to
keep the good stuff coming. So, following your approach, the way I should
proceed to "disprove" this model would be to change the environment in such a
way that tumbling when nutrients are declining no longer leads to better
nutrient rates, whereas tumbling when nutrients are increasing does. When the
UNALTERED organism-model fails to climb the nutrient gradient, this will
constitute proof that the model only works because of an "artifact" of the
environment. Your equivalent strategy is given below:

This doesn't cause a problem because (as Bill notes) a "chance property
of the situation" (what I call an "artifact") makes this logical path
least probable. When this "chance property of the situation" (the
fact that the gradient is more likely to be smaller after a tumble when
going up the gradient) is eliminated (as it is by my "de-artifacting"
code) the law of effect no longer "selects" the appropriate actions.

In other words, if we change the environment so that there is nothing to
learn, e. coli will learn nothing. Brilliant! Poof, the law of effect is
disproven!

Bill Powers (941125.1620 MST) to Bruce Abbott --

Have you tried to make your model work in a gradient of a repellant? As
it is stated, it clearly won't work: an increase in a repellant is not
reinforcing. You will have to change some definitions, particularly the
definitions of S+ and S-. Now a positive dNut must be considered an
unfavorable condition, and an increase in dNut from before to after a
tumble must be considered punishing. So the logic has to adjusted on the
basis that we know the organism will see a positive dNut as having
negative _value to the organism_. We must change our concept of the
organism's reference level for dNut.

I can't wait to hear the response to this; I've always wondered how the
law of effect knew which effect were good (positive) and which were
bad (negative) effects.

No problem--just reverse the signs in front of the learning speed parameter
within the learning procedure so that the repellant's effects are opposite to
those of the attractant.

Rick: I'm NOT going to repeat Thorndike's operational definitions for
satisfiers and annoyers a third time. The law of effect does not "know" which
effects are good and bad, so it must use an empirical test. Tell me, how does
a PCT researcher know what consequences are good and bad? It seems to me that
you must determine this by learning what perceptual variable is being
controlled, and what consequences of behavior will correct disturbances to it.
For example, before you can determine that, say, obtaining a piece of cheese
will serve as a lower-level system goal, you must first determine that
ingesting the cheese will help to correct the error in a higher-level system.
And that property--its ability to correct nutritional deficits--is partly
contained in the cheese, in its chemical constitution--a "dormative
principle," you might say.

Bill Powers (941126.0515 MST) to Bruce Abbott (941121.1830) and
Hans Blom (941123) --

Correct me if I'm wrong, but this implies that a particular consequence
of behavior, once it has determined what behavior is to be retained, is
always brought about by the same behavior.

Now I ask both of you: what would be the circumstances in which control
theory would predict that the same consequence will be produced by the
same action every time?

Great question.

Yes, and one that deserves a more extended answer than I have time for now.
For now I'll just note that Thorkdike's cats showed remarkably stereotyped
behavior in the puzzlebox, once they learned how to escape (see films by
Guthrie in his replication.). But you have the implication wrong. If all
swans are white, this does not imply that all white birds are swans.

Regards,

Bruce

<[Bill Leach 941127.19:20 EST(EDT)]

[Bruce Abbott (941127.1900)]

So what? This is what the environment sometimes "rewards."

I probably should stay out of this but it still seems that Rick is right
though the "proper" test would be a comparison with real E.Coli under
variable conditions. I do think that his point that humans attempting to
simulate e.coli behaviour behave in the same fashion as his model as
opposed to the "Law of Effect" model is significant.

So, following your approach, the way I should proceed to "disprove" this
model would be to change the environment in such a way that tumbling
when nutrients are declining no longer leads to better nutrient rates,
whereas tumbling when nutrients are increasing does. When the UNALTERED
organism-model fails to climb the nutrient gradient, this will
constitute proof that the model only works because of an "artifact" of
the environment.

What you are proposing is itself a control system operating in opposition
to the e.coli. What Rick did was to enter a random variation in the
concentration THAT ITSELF was not related to the e.coli's specific
action. Indeed, I think that he could further demonstrate his point by
making the concentration change occur at random times (totally unrelated
to e.coli tumbles) and still obtain similar results.

Though conjecture without modeling is not a real good idea... I would say
that unless you were to specifically design your test to make a control
sytem fail by controlling with more "force", the control system will
still work toward the goal.

In other words, if we change the environment so that there is nothing to
learn, e. coli will learn nothing. Brilliant! Poof, the law of effect
is disproven!

This is, of course, a valid statement though there is certainly an also
valid challenge to the claim that the "Law of Effect" model is learning
even when the environment is perfectly suited to it's parameter changes.

..., how does a PCT researcher know what consequences are good and bad?

By definition; those that result in the least perceptual/reference error
in the subject are "good".

It seems to me that you must determine this by learning what perceptual
variable is being controlled, and what consequences of behavior will
correct disturbances to it.

This is actually a pretty good statement (in my opinion) though it
potentially still might cause one to think in terms of consequences
controlling... as seems to be implied below:

For example, before you can determine that, say, obtaining a piece of
cheese will serve as a lower-level system goal, you must first determine
that ingesting the cheese will help to correct the error in a
higher-level system. And that property--its ability to correct
nutritional deficits--is partly contained in the cheese, in its chemical
constitution--a "dormative principle," you might say.

This paragraph seems to be full of implications. In the first place, I
know of few examples where specific nutritional deficiencies have been
show to result in a control action to result in a correction of such
deficiencies. You are implying that I (for example -- one that happens
to be a great cheese lover), eat cheese to satisfy some nutritional
deficiency.

Why did I just now go and get some pretzels and begin eating them? Do I
have some nutritional deficiency satisfied by pretzels and thus the
consequence (eating pretzels) "controlled" my actions?

I think not... in fact I would rather have the cheeze but it is more of a
bother and a little tougher to eat while typing at a computer keyboard.
If the pretzels were not available, I would have pursued some other
option.

If I am hungry and you control (by force) ALL of my sources to food then
you might be ALMOST justified in a claim that your enforced consequences
are "controlling" my behaviour. If you are not controlling ALL of my
food sources then I may choose to comply with your manipulations or I may
not. If I do choose to comply then to the extent that your manipulations
control the physical reality that I must deal with you are "controlling"
but I think that even this is badly misleading.

... For now I'll just note that Thorkdike's cats showed remarkably
stereotyped behavior in the puzzlebox, once they learned how to escape
But you have the implication wrong.

Is not Pavlov's Dog an example of "operant conditioning"?

As to the question... If one means by "action"; behaviour or output then
PCT predicts that such will occur when the same set of relevent
perceptual references exist and the same disturbance is applied to the
same physical environment.

I would suggest that this is a bit of an over-simplification but not
much. In the case of the types of experiments so far discussed previous
experience (learning) can greatly affect specific behaviour.

-bill