Re: The animal, not the environment, controls behavior.

JGCASEY wrote:
Michael Olea wrote:
JGCASEY wrote:

Seeing does involve behavior as the eyes are moved about
to collect information and this may be guided by innate
and learned stimulus responses. But I see seeing as the
use of that collected information.

This is jonesing for a question - what is "the use of
collected information"? What information comes to be
collected, and how does this vary with experience?

The animal collects information that is useful to the animal
because the animal, not the environment, finds the information
collecting and the information collected as rewarding. The
animal, not the environment, uses that information to seek
further rewards. Without those changes in the animal nothing
happens.

I get the feeling here that you are responding to something other than what
I wrote. Did I imply that the environment finds collecting information
rewarding? What the environment does is create conditions that have
consequences. Or perhaps I should say the environment *is* a set of
conditions, which, by the way, have consequences. In any case, information
is useful to an organism if it predicts consequences of relevance to the
organism. Now, you could argue that I've simply pushed the term "useful"
into the term "relevant", but the key term is "predicts". Not all
information is "predictive information". Not all features of the
environment are equally "discriminative" (which, by the way, is why feature
selection is an important part of designing a classifier).

Maybe my question was too glib. I was asking you to clarify what you meant
by "I see seeing as the use of that collected information".

It is the animal, not the enviromental, that decides what visual
cues to use to control eye movements along with any innate
determinates.

Environments don't "decide" anything, they contain clues that signal likely
consequences. Consider saccades. You make, on average, 3 of them per
second. Can you describe on what basis you decide what to fixate, and in
what order, and for how long?

http://journalofvision.org/4/8/737/
http://journalofvision.org/5/8/917/
http://books.nips.cc/papers/files/nips18/NIPS2005_0199.pdf

http://citeseer.ist.psu.edu/hommel03planning.html

We also use our expectations and assumptions about
spatial relationships to decide where to look next.

Yes, and those expectations are, by and large, learned
- they have come under "stimulus control".

But it is not a one way control situation.

"I have often speculated as to how well James Gleick's best-seller would
have fared at the bookstores if it had borne a title like 'Sensitive
Dependence: Making a New Science'".
-- Edward Lorenz, "The Essence of Chaos"

I suspect the phrase "stimulus control" tends to cloud judgement: "I'm not
under the control of no stinking stimulus". Meanwhile, observe a busy
intersection with a traffic light. Or go to the lobby of a movie theater,
and observe which restrooms people enter.

A feature of the environment acts as a stimulus control if it is predictive
of the likely consequences of behavior - as is a traffic light at a busy
intersection. I believe the term "control" here is in reference to the
distinction between an independent variable and a dependent variable. In
the traffic light example there is a correlation between stimulus - the
state of the traffic light - and the behavior of motorists. The information
is mutual, given one you can predict the other. So which is the dependent
variable and which the independent variable? Try an experiment - drive
through an intersection and see if the light changes. Now change the state
of the light and see if the behavior of motorists changes. Or walk into a
restroom marked "women" and see if the sign changes to "men". Now switch
the signs on the restrooms and see if there is a change in restroom
entering behavior. Which are the dependent and which the independent
variables? What is, statistically, controlling what?

The environment alone does not predict the animal behavior.

Who said it did? For one thing, reinforcement is relative. Here is how I
tried to explain it before:

=====

A "food deprived" rat places a greater value on access to food (will work
for food) than does a satiated one. But, you can deprive a rat of the
opportunity to run on its running wheel (say the wheel is locked in place),
and use opportunities to run to reinforce eating responses (I think parents
do something like that with their kids). In this case running acts as a
reinforcer of eating. Or you can restrict a rat's opportunities to eat, and
use eating to reinforce running (I think society does something like that
with its citizens). ...

I could not help but think about this when I read comments made by the
reporter that had been kidnapped in Iraq, and released much later. I forget
the exact quotes, but she spoke of the joy of being able to go outside, see
the light streaming through the trees, feel the breeze on her face.

=====

And like this:

=====

Suppose you are driving somewhere and approaching an intersection with a
traffic light. You can, to a first approximation, drive through the
intersection, or stop before entering it (along with many variants,
including slowing down, maintaining, or increasing speed as you approach).
Among the possible consequences are: getting in an accident in the
intersection, getting a ticket, passing through the intersection without
incident and perhaps arriving more quickly at your destination, being
rear-ended... The probability distribution over the consequences is not
independent of your response, nor is it independent of the state of the
traffic light. The state of the light is not the only context acting as a
stimulus control - there is the time of day, the ammount of traffic, the
visibility to either side, whether or not you hear a siren (ambulance,
police, or fire truck, not the Odyssean variety). Your probable response
will not be independent of the probability distribution over the
consequences given the context, even though it is quite unlikely you can
compute the distribution. It will also depend on the urgency of the task,
the relative gain ratios over the consequences. Are you sight seeing, going
to shop for groceries, late for a movie, late for an important meeting at
work, rushing to the hospital, fleeing speeding gunmen firing at you,
trying to catch Angelina Jolie before she gets on a plane and departs your
life forever?

Odds are you cannot calculate the odds, yet the probability distribution
over your responses and most people's responses will reflect those odds.
Usually you will stop at a red light ... unless Angelina might get away.

======

Each animal, in the same environment, will behave and "come
under stimulus control" according to its internal makeup.

And its history of interaction with the environment. Take, for example, the
features of auditory stimuli to which speakers of different languages
attend.

Sure you can see what parts of the environment the *animal
chooses to respond to* and then claim those parts of the
environment are controlling the animal.

And I can, within limits, systematicaly vary the parts of the environment
the animal "chooses to respond to" by systematicaly varying the parts of the
environment that signal a correlation between action and consequence. The
features we come to pay attention to are those that have predictive power.
We don't, in general, get to choose what features those will be any more
than we get to choose the temperature of the sun. You might argue that we
can choose what it is we want to predict, but we do not, in general, get to
choose what features of the environment predict what we choose to predict.

But then you have
to explain why environment e excerts one kind of control
on animal rat and another kind of control on animal pigeon.

No, I don't. No one is arguing against species-specific behavior, or that
different species attend preferentialy to different features of stimuli.
However, the *form* of the relationship between stimulus, behavior, and
probable consequences is largely the same for rats and pigeons, even though
the specifc stimuli, behaviors, and consequences differ. And rats, pigeons,
and people attend preferentialy to stimulus features that predict
consequences.

You can claim that the dog is under "stimulus control" of
the whistle or you can claim that the dog has internally
associated the whistle with an internally determined reward.

What are the testable differences between these claims? Suppose the whistle,
somehow, were to lose any correlation with consequences - how long would
the dog continue to "choose" to attend to whistles? What experimental work
along these lines has been done?

If an animal seeks or avoids a particular stimulus it is the
animal, not the particular stimulus, that is determing the
behavior given a particular stimulus. The animal makes that
association and therefore is best seen as the controller.

How does this view improve the predictability of the relations between
stimuli and behavior?

-- Michael


.

Relevant Pages