PCT vs conventional approaches

[From Rick Marken (960921.0920)]

Rupert Young (960920 1100 BST) --

Welcome to CSGNet, Rupert.

Bill Powers (960920.0900 MDT) answered two of your three questions; I'll
take a shot at the one he didn't answer:

a) How does HPCT's approach to animal systems, of hierarchical levels of
control, conflict with conventional psychology ?

The basic conflict is captured by the title of Bill's book: _Behavior: The
control of perception_. Conventional psychology is based on the idea that
behavior (actions and the results of those actions) is ultimately caused by
perceptual inputs. The title of a book about the conventional psychology
approach to animal systems could be: _Perception: The cause of behavior_.
(Some conventional psychologists, like J. B. Watson and B. F. Skinner,
might take an even stronger stance and call their book _Perception: The
control of behavior_. In this case it is the perception of a _reinforcer_
that controls behavior).

Even modern cognitive science and complex systems approaches to behavior
are based on the idea that behavior is _caused_, though not necessarily by
perception. In cognitive science the causes of behavior are "goals" or
"plans" in the brain. In the complex systems approach, behavior is defined
by differential equations that describe causal relationships between many
variables; the equilibrium state of this system of equations is the "end"
or "goal" of behavior: behavior is again the end result of a causal
process.

In PCT, behavior is not an effect that is caused. Rather, behavior is a
process: the process of _controlling_. PCT starts with the observation that
the term "behavior" is ambiguous; it refers to both actions and the results
of those actions; to flapping wings (action) and flying through the air
(result). PCT also recognizes that organisms are not the only cause of the
results of their actions; the results of action are also caused by
environmental variables (disturbances) that are quite unpredictable. For
example, flying is caused not only by flapping wings but also by air moving
at a particular speed and angle relative to the wing. Despite these
disturbances, organisms act to produce consistent results; the bird flies
despite continuous variation in the speed and angle of the wind relative to
its wings. Organisms do this by adjusting their actions _as necessary_ to
produce the intended result; the bird adjusts the flapping of its wings to
produce exactly the lift needed to fly. The process of producing consistent
(intended) results in the context of variable disturbances is called
_control_. It is also called _purposeful behavior_. Organisms control by
acting to bring a perceptual representation of the intended result into a
match with an internal (to the organism) reference specification for that
result. Behavior is the control of perception.

The process of control is not a cause-effect process; control occurs in a
loop where every variable is both cause and effect at the same time.
Nevertheless, control can _look like_ a cause-effect process when viewed
from outside the behaving system (see my paper "The Blind Men and the
Elephant" which is available on the Web via the CSGNet References section).
Basically, the mutual (and necessarily correlated) effect of disturbance
and action on an intended result (controlled variable) gives the illusion
that external events (distubances) cause responses (actions). The
conventional approach to understanding behavior is based on the assumption
that this disturbance-action relationship reflects the existence of a
causal path through the organism, from environment to perception to
"behavior" (response or action). The PCT approach to behavior shows that
the existence of this causal path is an illusion. This is the basic
conflict between the conventional and PCT approaches to behavior.
Unfortunately, it is not a conflict that can be resolved by compromise; if
organisms are control systems, then the conventional approach to
understanding behavior "has been in the grip of a powerful illusion since
it's conceptual bases were laid" (Powers, Science, 1973). This is
obviously something, up with which conventional psychologists will not put;
hence, the overwhelming lack of interest in PCT since the publication of
_Behavior: The control of perception_.

What all this means for HPCT is that hierarchical models based on
cause-effect principles (Albus has one; Brooks at least talks about his
this way) are fundamentally different from HPCT; they are based on the
hierarchical generation (cause) of output (actions). HPCT is based on the
hierarchical control of input (perception). To get an idea of how an HPCT
model work, see my paper _Spreadsheet analysis of a hierarchical control
system model of behavior_ which is available on the Web as one of the
chapters in my _Mind Readings_ book.

Best

Rick

[From Rupert Young (960924.1700 BST)}

(Rick Marken (960921.0920)

In PCT, behavior is not an effect that is caused. Rather, behavior is a
process: the process of _controlling_. ...

.... Organisms control by
acting to bring a perceptual representation of the intended result into a
match with an internal (to the organism) reference specification for that
result. Behavior is the control of perception.

Frogs, apparently, whenever they see a small, fast-moving, buzzy thing stick out
their tongues in an automatic, hard-wired fashion to eat the fly. This seems a
clear cut case of stimulus-response/cause-effect behaviour. I'm not sure how
this fits into PCT. What is the "perceptual representation of the intended
result" and where is there a control loop ?

To get an idea of how an HPCT
model work, see my paper _Spreadsheet analysis of a hierarchical control
system model of behavior_ which is available on the Web as one of the
chapters in my _Mind Readings_ book.

I will do once I can get the demos sorted out.
Looks interesting and just what I'm looking for.

Cheers
Rupert

[Avery Andrews 960926]
  (Rupert Young (960924.1700 BST))

>Frogs, apparently, whenever they see a small, fast-moving, buzzy thing stick out
>their tongues in an automatic, hard-wired fashion to eat the fly. This seems a
>clear cut case of stimulus-response/cause-effect behaviour. I'm not sure how
>this fits into PCT. What is the "perceptual representation of the intended
>result" and where is there a control loop ?

Well, one possibility is a loop whose reference is `no bugs', where the
activity induced by the error signal is to track the bug and eat it (they
certainly do track the bugs, and get into an optimal orientation for the
toungue-strike to succeed). `Apparently' is another key word here; I've
found that rather often, in the sections of papers where people talk about
the limitations and inadequacies of closed-loop models, the discussion is
some combination of flimsy, empirically obsolete or very badly
thought-through. For example, has anyone investigated whether there's any
effect of hunger on the `avidity' with which frogs clean up bugs? Many herps
certainly feed more eagerly when they haven't eaten recently than when they
have. This is very visible for, say Australian bluetongue-lizards, but it's
harder to tell with frogs, so I really don't know.

The real point of this story is not whether there is or isn't a good closed-loop
account of this particular piece of behavior (or any other), but that it's
important not to give up too easily in the search for one. The most resistant
case we've encountered here, where I think Rick Marken and Bill Powers may
have more or less given up, is a moth that, when it hears bat-sonar, folds
its wings and plummets to the ground. This is hard to construe as control
because the wing-folding doesn't make the sonar signal go away. On the other
hand, there's the question of how this `response' gets integrated into the
rest of the moth's control systems, maybe it will turn out to be best to
construe this behavior as the result of a control-system whose reference is:
`no wing-flapping when sonar is audible' (see Rick's spreadsheets for a demo
with a bit of this kind of logical control in it).

  Avery.Andrews@anu.edu.au

[From Bill Powers (960926.0530 MDT)]

Avery Andrews 960926]
  (Rupert Young (960924.1700 BST))--

Avery says, re Rupert's example of the frog's mode of catching flies,

The real point of this story is not whether there is or isn't a good
closed-loop account of this particular piece of behavior (or any other),

but >that it's important not to give up too easily in the search for one.

Along with this important point, which I'm grateful to Avery for mentioning,
is its corrolary: it's important not to accept any explanation, even the
conventional one, just because it seems to support what you believe. What
"What the frog's eye tells the frog's brain" tells us is mainly that frogs
can't reorganize their motor control systems after the eye is surgically
inverted. Nothing in this study tells us how the normal system is actually
organized.

There are many examples of apparently open-loop behavior: frogs and
chameleons throwing their tongues at flies, mantises striking at prey;
chicks pecking at pieces of grain, moths (as Avery mentions) folding their
wings and plummeting to the ground, human subjects extending a finger as
rapidly as possible toward a target, human beings playing the piano very
rapidly, and even (my example) human beings bowling. Nobody seems to bring
up such examples on this net as the beginning of an effort to devise a
control model; it's more often a case of "Oh yeah, well what about THIS
example?" The person offering such examples seldom follows them with an
attempt to find a control model that might explain them.

My impression is that in all these studies (except the last), one overt or
covert goal was (at the time they were done) to show that an SR analysis was
sufficient and there was no need to invoke the idea of purposive controlled
behavior. The history of this kind of experiment goes all the way back to
the days when scientists were still arguing about purpose and long before
they were even dimly aware that they were arguing about control. Since most
scientists have believed that behavior is caused by input stimuli, no
serious attempt was made to examine the situation for aspects that might
demand some other explanation.

I've always been puzzled, in re the frog, why nobody found it remarkable
that the lightening dart of the tongue comes anywhere near the target,
especially a moving one. The speed of the action obviously makes motion less
important, because the target can't move much during the rapid extension of
the tongue, but how is it that the tongue is aimed so accurately that its
sticky tip arrives at a place four or five inches away with an accuracy
measured by the combined diameters of the tip of the tongue and of the fly?
The process is much like aiming a gun at a target and pulling the trigger.
Why does everyone want to ignore the aiming phase of this process? Because
it seems too purposeful?

When a golfer swings at a ball, the golfing pro will say that it's important
to create a consistent swing, to remain in balance, and above all to keep
the head still. Tiger Woods, who is a very good young golfer, keeps his head
still with uncanny accuracy; he reminds me of those birds whose heads snap
from one spatial location to the next while the walking body moves
continuously along underneath, or a bird landing on a telephone wire and
holding its head in one place while the body, now attached to the wire,
swings gently back and forth.

Apparently, for performing rapid actions where control would be next to
impossible during the action, it's important to maintain a stable base so
that the aiming point will reliably determine where the sudden action will
go. If the relation of the output act to the supporting body and aiming eye
varies, then so will the result vary. When the prey can detect approaching
danger and start moving in an unpredictable direction to avoid it, it's
necessary to make the act as rapid as possible; the more rapid the action,
the more important it becomes to be able to hit a selected point whether or
not anything is still there at the end of the action. For the frog, I
suspect that the main thing is to orient the prey at a location relative to
the head and body such that the tongue will arrive at the same visual
position where the prey is at the moment the action starts. Just like
putting the target on the cross-hairs before pulling the trigger.

The main question that is answered by inverting the frog's eye or putting
prism goggles on chicks is whether, when the relation between the visual
image and the body is altered, the system can reorganize after enough
experience with missing the target to start hitting the target again. For
some animals the answer is no; for others (and for slower actions) it is
yes. But even in the cases where the answer seems to be no, I wonder what a
Skinnerian would say -- would he admit, for example, that he couldn't train
a chicken to peck two inches to the right of a piece of grain in order to
get fed? Do we know that there isn't some range of variation of the frog's
eye position that the frog could learn to compensate? Inverting the eye is a
pretty huge disturbance!

The main question is how much proof is required to satisfy you that your
initial idea was wrong, as opposed to how much is required to show you that
you were right. I think it's natural for human beings to accept confirming
evidence or arguments more readily than disconfirming evidence or arguments,
but that doesn't mean that good science can be done that way. All the best
scientists I know (some of whom have PhDs) find confirming evidence harder
to believe than disconfirming evidence; when an experiment goes right, they
immediately start searching for what they did wrong. Unfortunately, some
scientists who seek public recognition (even some who have PhDs) take
confirming evidence as the occasion to relax their strict requirements of
proof, which they reserve mainly for their rivals' work.

Best to all,

Bill P.

[From Bruce Gregory (960926.1045 EDT)]

(Bill Powers 960926.0530 MDT)

There are many examples of apparently open-loop behavior: frogs and
chameleons throwing their tongues at flies, mantises striking at prey;
chicks pecking at pieces of grain, moths (as Avery mentions) folding their
wings and plummeting to the ground, human subjects extending a finger as
rapidly as possible toward a target, human beings playing the piano very
rapidly, and even (my example) human beings bowling. Nobody seems to bring
up such examples on this net as the beginning of an effort to devise a
control model; it's more often a case of "Oh yeah, well what about THIS
example?" The person offering such examples seldom follows them with an
attempt to find a control model that might explain them.

This is my feeling too. I am reminded of exchanges in which one
side brings up examples to challenge an adaptationist
interpretation of evolutionary change and the other side
invents adaptationist explanations of the phenomenon. These
shoot-outs no doubt have some value, but as far as I know, no
one has ever changed his or her viewpoint as a result. I would
much prefer to see a concerted effort to push PCT explanations
as far as they can go. Of course, we could always adopt the
ground-rule that no one bring up an apparent example of
open-loop behavior unless he or she does so in order to propose
a closed-loop explanation...

Bruce