[From Bill Powers (2009.07.31.1636 MDT)]
Shannon Williams (2009.07.30.1900 CST) –
[From Bill Powers
(2009.07.30.1040 MDT)]
But you can’t mean "if I produce the same actions every time I
am more
likely to survive," because that is false.
Can you tell from my response to Rick what I mean to say?
Imagine
yourself in an environment that you are unfamiliar with. Imagine
the
first couple of times that you perform some task. To get real
extensive you can imagine that the environment contains garroters or
rapists or lions or tigers or bears. And you have to get from point
A
to the food at point B. Or you can pretend that your food is
at
point A with you. Your survival is ensured simply if you never
leave
point A. This, though, gets more into the second statement that
I
asked of Rick.
There are problems with what I guess is your proposed answer, and they
come from the assumption that cause and effect always work the same way:
that is, the assumption that if action A is performed and has result B,
performing action A again will result again in B. When you come across
this sort of idea in science, specifically behaviorism, there is always,
in the background, a disclaimer: “CETERUS PARIBUS” or
“All else being the same.” The trouble is that in the real
world all else is never the same. If sheltering under a tree to escape a
storm works one time, doing that again will probably get you electrocuted
or drowned if the storm comes from a different direction.
If you’re very careful and have control of all the details, you can set
up an experiment in which event A will, indeed, quite often but
unfortunately not always, lead to event B. But out in the real world
that isn’t possible. B. F. Skinner discovered this by observing rats
pressing levers in cages. To be sure, if the lever was depressed all the
way, a pellet of food would be delivered for every Nth press, where N was
determined by the experimenter. And rats could learn to keep pressing
and, on every Nth depression of the lever, get a pellet and eat
it.
But what behavior, what action, were they learning? Skinner thought at
first that they were learning to generate repetitive movements that would
reliably cause the lever to move downward. When I heard of that I thought
of a rat sitting beside the lever and pushing down again and again with
one paw. But Skinner found, and I realized after seeing videos of this
sort of behavior, that the rats were learning, sort of, only that the
lever needed to move downward to a limit. How they accomplished that
result was extremely variable. Sometimes they would press the lever with
one paw just I imagined. But sometimes they would press it with the other
front paw or one of the back ones, or bite the lever and make it go down,
or sit on it – they did whatever actual behavior was handy from where
they were to get the lever to go down, and it was far from being the same
behavior every time. A lot of the time, lever presses seem to be simply
accidents; the rat would bump into the lever while trying to do something
else, like get its nose into the cup where the pellets appear (if it’s
close to the lever).
Skinner handled this problem by defining “the Operant.” The
Operant is the class of all behaviors that have the same consequence. The
Operant called “lever pressing” is made up of all the different
– vastly different – behaviors that end up with the lever being
depressed far enough for a press to be counted (in videos you see many
presses that don’t quite get that far). This creates enormous problems
for any causal explanation of behavior, but Skinner simply ignored that
problem – he wasn’t interested in how the right behavior was
generated. That was a mistake.
This same problem also exists even if you say that A is a controlled
result of behavior. The behavior required to produce situation A may be
different every time A is produced, but you can overlook that problem if
you simply say that every time A happens (for whatever reason), B
happens.
You can say, in fact, “Every time you do A, B happens.” The
word “do” indicates vaguely that some unnamed action of yours
caused A to happen, which in turn caused B to happen. If a different
behavior is needed to produce A, you can still say you’re
“doing” A because “do” doesn’t specify how you
“do” it. The untidy fact that the action has to be slightly or
greatly different every time just gets swept under the rug.
What is it that comes under stimulus control? Pushing with the any one of
the four paws, biting or nosing or climbing over the lever, and sitting
on it – whichever one of those, or any other actions, makes the lever
move. Somehow the discriminative stimulus knows, before each lever press,
which action needs to be chosen to get the lever to depress this time. In
truth, what is being measured by the experimenter is only the fact that
the lever does go downward. The behavior that causes each depression goes
unnoted. Yet the discriminative stimulus, it is said, causes the lever to
be pressed. That is not an explanation or a theory; it’s a serious causal
paradox that needs to be solved. PCT, of course, solves it.
Shannon, you seem to be working toward saying that we learn to do what
works in a given environment. But how do we learn that, when in a real
environment we must normally alter our actions to make the same effect
occur again? We do whatever it takes to make A happen, because from
experience we know that if A happens, we can expect what we want to
happen, which is B. What kind of model of behavioral organization can do
that? Not the causal models historically behind psychology. The only
behavioral organization I know of that can do that is a negative feedback
control system, the central idea of PCT.
My answer is quite like the one I assume you want. We learn to do
whatever it takes to make something be a certain way, whether it’s that
something we’re interested in or some subsequent reasonably reliable
effect of that something. I haven’t been interested in what things we
need to make happen, but in how an organism can learn what works, and why
it learns that.
The how and why are wrapped up in reorganization theory. The why is
simply because we have lost control of something vital and either can’t
get what we need or can’t avoid something that harms us. We need to learn
something new.
The “how” part is what psychologists have been calling
“trial and error” learning. By “trial” they mean an
organism varying the behavior it produces to see what will happen, and by
“error” they mean that if what happens is no better or is
worse, try something else. At random. When you get down to trial and
error, that’s because nothing else you know how to do has worked. Trial
and error is the only way to find something NEW to do. Do something
different – unlike anything else you can think of. That’s what random
means.
The particular algorithm I use for reorganization is modeled after the
way E. coli bacteria manage to swim up gradients of attractants or down
gradients of repellents. It works very nicely as far as I’ve been able to
see in simulations. I can’t say yet how it’s carried out in real
organisms, but I’ve been seeing hints in Science and in Nature that
random elements are beginning to show up in biochemical systems,
particular processes of gene expression in DNA. There’s a lot of evidence
from many places that when there is stress, behavior and gene expression
become more random. The more cell biologist learn, apparently, the less
sure they are getting about the simple lineal concept of causation or
determinism. I don’t know if they’re about to discover PCT reorganization
theory or something else, but it’s nice to know they’re starting to lean
our way.
So, Shannon, why not just come out with your picture? Actually,
scientists usually prefer for the other guy to state his entire case
before offering criticisms or approval, because they don’t like to offer
the opportunity for the other guy to quickly change what he was going to
say to avoid a criticism he can’t counter. Or she. It’s best to avoid
even the suspicion of doing that,
Best,
Bill P.