Learning proposal

[From Bruce Abbott (961126.1215 EST)]

Rick Marken (961125.1600) --

Bruce Abbott (961125.1540 EST)

Actions are movements, acts are perceived consequences of movements. An act
might be a circle, drawn in the air by the tip of my finger as a result of a
certain sequence of actions (muscle contractions)...The act is a perception
arising from behavior.

A particular act may be intended or unintended.

This is an interesting new concept -- an unintented perceived consequence of
actions. Ok. So learning (according to your model) involves changing
perceived consequences of action from being unintended to being intended.
This means developing a reference for some state of these perceptions.

Hey, an "interesting new concept." Now we're getting somewhere! Yes,
exactly: this means developing a reference for some state of these perceptions.

When the rat first depressed the lever, that act was probably a side-effect
of activity related to controlling some other variable. When that act is
followed by the delivery of a food pellet, the rat tends to repeat the
series of acts that immediately preceded that delivery.

Interesting. So the rat has some memory of all the perceived consequences of
its actions -- intended and unintended -- that happened "prior" to the
reward.

That's right, Rick: I am assuming that the rat remembers what was doing when
the pellet appeared. You're doing great!

And then, for some reason, after the reward the rat "tends" to repeat
a "series of acts" that immediately preceded that delivery.

The number of perceived consequences of actions (acts) that precede (by how
much?) a reward is very large - - probably equivalent to the number of
afferent neurons in the NS. Which acts or series of acts does the rat _tend_
to repeat?

Before continuing, I want to make it clear that I include the sensory
feedback produced by those movements (including tactile and proprioceptive
ones) among the "perceived consequences" of actions. Which acts does the
rat tend to repeat? Excellent question. I would hypothesize that these are
the ones that the rat was doing just prior to the appearance of the pellet.

Why does the rat "tend" to repeat them?

I used the word "tend" because the rat is not always observed to repeat
these acts -- quite possibly because other control systems (e.g., those
producing exploratory behavior) sometimes get asserted at the time when this
repetition of the acts should be getting underway. For theoretical
purposes, we can assume that these acts _will_ be repeated, under the
assumption that there are no conflicts with other systems getting in the way.

How strong is the tendency
to repeat them. If a reward occurs _while_ the rat is repeating these acts,
does the rat stop repeating the acts and then restart again? If so, at what
point?

Great questions. I believe the proposal would say that if a reward occurs
_while_ the rat is repeating these acts, that they would be cut short; now
that the pellet is available, the portion of the sequence beginning with
approaching the pellet and ending with its consumption would commence. When
the pellet was gone, the rat would then repeat what it was doing previously
when the pellet appeared.

How does it know about the beginning and end of "acts" and "rewards".

The end of the acts is defined by pellet delivery. How the beginning would
be defined is a matter for research, but it is no doubt related to the
duration over which the memory of the acts prior to pellet delivery
persists. Appearance of the pellet is what the currently active control
system is "looking for." The appearance and disappearance of the pellet is
a simple matter of perception.

Your model has more loose ends than I originally suspected.

Yes, it does have a number of gaps -- I've only begun to develop it. It
requires that I suggest specific mechanisms, such as one to activate the
"replay" of the perceptual memory for recent acts -- which would act as the
reference variables for reproducing the remembered acts via lower control
systems.

What was a side-effect of activity directed toward other goals

What does this mean? The side-effect (non-goal directed) activity was
directed toward goals all along?

No. An example may clarify: Depressing the lever occurred not because the
rat wanted to press the lever, but because, e.g., it wanted to rear up and
sniff the upper part of the chamber. It placed its paw against the lever
and, pushing down in order to raise itself up, as a side-effect the rat
tripped the switch.

now becomes the goal itself -- the rat intends to repeat that perceived
pattern of activity, i.e., what it was _doing_ on the previous occasion just
before the pellet was delivered.

OK. So the rat ends up repeating (for some unspecified reason and by some
unspecified means) one perceived consequence of its actions -- say, the
perception of the lever in a particular state. But what if putting the
level in a particular state was not the repeated act? Suppose the
rat had repeated the tongue movement perception it produced just before
the reward. How does the rat select new perceptual consequences to repeat?

Good question. It could try something else at random, though I doubt this
would be the case. More likely it will return to exploration of the chamber
("looking for food"), which may eventually lead to another tripping of the
lever. This is probably the default mode what might be called the "food
acquisition" control system.

Your model assumes that learning leads to the particular act or act stream
(perception of the lever in a particular state) that produces reward. But it
is often the case that quite different acts -- ones that have never before
been followed by reward -- must be produced in order to produce the reward.

Yes, and my proposal allows those previously unrewarded acts to occur.

The "acts" you describe are actually perceptual _variables_.

Yes, but they are very special ones: they are under the animal's control.

Under most
circumstances, organisms can control a reward only by varying the reference
state to which the "act" perception is brought.

That is how the act is voluntarily produced.

This can be demonstrated in a
tracking task where subjects must (and DO) learn how to vary their reference
for the kinesthetic perception of handle position in order to get the reward
of having the cursor stay in the intended location.

Yes, it's called "trial and error" learning. But these are not random
changes; one tries slight variations, looking for improvement in terms of
minimal cursor movement, reduction in overshoot, etc. What goes on is, I
think, more sophisticated than e-coli type random reorganization.

This is basic hierarchical control theory. I suggest that you carefully
reread my "Spreadsheet hierarchy..." paper (and play with the spreadsheet
model a bit) to see how it works.

I know how it works, I've played with the model, and it's a nice
demonstration. But I doubt that it works efficiently enough in the enormous
parameter space in which living organisims exist. Something vastly more
efficient is required there than random reorganization of parameters, in my
opinion, and my little story, vague as it is in some important details, at
least provides a starting point on the way toward an alternative.

I'd like you to try the following and give my your thoughts:

Demonstration 1
  Move your arm so that your finger tip traces out a circle in the air.
  Now that you've done it, can you remember what it looked like? Felt like?
  Can you reproduce the act? If so, how?

Demonstration 2
  Watch someone else perform some simple act (e.g., picking up a coffee cup).
  Now reproduce that act (i.e., do it yourself). Were you able to? How did
  you do that?

Regards,

Bruce

[From Bruce Abbott (961126.0900 EST)]

Rick Marken (961126.1330) --

Bruce Abbott (961126.1215 EST)

That is how the act is voluntarily produced.

But that doesn't explain how the animal learns to control. If you ever
actually develop a working version of this model, you will find that the
model, as you describe it, cannot possibly learn to control -- a big failing
in a model of learning;-)

I think I'm hearing wishful opinion; certainly I don't find any scrap of
reasoning to back it up. It won't work because Rick Marken says it won't,
that about it?

I'd like you to try the following and give my your thoughts:

Demonstration 1
Move your arm so that your finger tip traces out a circle in the air.
Now that you've done it, can you remember what it looked like? Felt like?
Can you reproduce the act?

Yes.

If so, how?

Got me. I think it has something to do with my ability to select remembered
reference specifications for perceptions.

How do you select remembered reference specifications for perceptions?

Demonstration 2
Watch someone else perform some simple act (e.g., picking up a coffee cup).
Now reproduce that act (i.e., do it yourself). Were you able to?

Yes.

How did you do that?

Again, it was probably by selection of remembered reference specifications
but, in this case, the selection was based on what I _imagined_ the
perceptual consequences of controlling these perceptions would be _for
another person_, one who would be in the same position watching me as I was
when watching the person I am now trying to imitate.

How were you able to imagine what the perceptual consequences of controlling
these perceptions would be for another person? How were you able to turn
these into reference specifications for exactly the right control systems?

I think these demonstrations indirectly suggest just how impossible it would
be for a rat to know which perceptions (acts) to repeat after receiving a
reward. In these demonstrations I knew, in advance, which perception I was to
repeat. The rat has no idea when it is going to get a reward so it has no
idea what perception(s) to repeat when it finally actually does get one.

I'm hypothesizing that it remembers what it was doing when the pellet
appeared and uses the remembered reference specifications for those acts to
repeat those acts. That severely limits the number of perceptions that need
to be repeated. Also, I think that the perception of "doing" something
probably doesn't normally include the thousands of parallel control systems
involved in keeping physiological variables within normal limits, so that
would impose a further restriction on the possibilities.

A reward occurs at some point during hundreds (thousands?) of continuous
parallel streams of controlled and uncontrolled perceptions: the organism is
continuously perceiving and controlling many visual, kinesthetic, visceral,
auditory, etc perceptions before and during the occurance of the reward. Your
model has the rat selecting, from all these perceptions (controlled and
uncontrolled) one or a "series" of several percpetions to repeat. The chance
that the organism will repeat just _the_ one perception (or series of
perceptions) that actually has anything to do with the occurance of the
reward is ridiculously small.

I doubt it it's at all that complex. If I drop my pen on the desk and it
happens to make a neat sound, I don't perceive having any trouble
identifying which perceptions I have to repeat so I can hear that sound
again. I doubt that the rat has any more trouble than I do.

I think this model of yours is based (surprise) on an old S-R proclivity to
perceive the world only at the event level. There are stimulus events (lights
going on or off, levers appearing or disappearing), response events (presses,
licks, pecks) and reinforcing events (food pellets, shocks). It's a nice,
discrete world where there is a place for everything and everything is in its
place. It's a world that exists at only one level of the 9+ level PCT
hierarchy.

Most things we do require that we complete one sequence (bring one
perception to reference) before we initiate the next. I've got to pick up
my shoe before I can put it on, stand up before I can walk, walk to the door
before I can open it. The rat must approach the lever before it can press
it, approach the food cup before it can sieze the food, place the food in
its mouth before it can chew and swallow the food. These are not
requirements of the nice, neat, artificial world of the operant chamber but
of the world in which all of us live, all of the time. To control some
variables effectively, one must organize a sequence correctly. My
hypothesis is intended to deal with that process, so naturally, it focuses
on these discrete events that signal the successful completion of one
control process and the initiation of the next.

When you eventually decide to leave the dark side and go all the way with the
brilliance of PCT, you'll realize that the world we experience is a world of
many different _types_ of continuously changing perceptual _variables_.

Yes, I know all about continuously changing perceptual variables: I model
them all the time. But I am really amazed at your world. It is a world
where all the difficult problems (like learning) have already been solved, a
bright, changeless, pristine world in which no theoretical concept will ever
need revision, so perfect are they in conception, so wonderous to behold.
Not at all as it is over here on the Dark Side, where mere mortals must
struggle to understand the nature of underlying reality from vague shadows
cast upon the wall.

Tell me about some of the empirical observations you see as offering
difficulties for HPCT as now constructed.

Regards,

Bruce

[From Rick Marken (961127.0840)]

Me:

you will find that the model, as you describe it, cannot possibly learn to
control -- a big failing >in a model of learning;-)

Bruce Abbott (961126.0900 EST) --

I think I'm hearing wishful opinion; certainly I don't find any scrap of
reasoning to back it up. It won't work because Rick Marken says it won't,
that about it?

It seems like cruelty to nightcrawlers to throw any more light on the dark
side than has already being thrown by Jedi knight sublime "Luke" Powers.
After all, when the force is with you, you know that people who actively
shield themselves from the light are _controlling_ for being in the dark.

But let me quickly explain why your learning model will not learn to control.
The basic problem (among MANY) is that your model, as described, simply
causes a reference to be set for the _specific perception_ (act) that
occurred prior to the reward.

Ignoring the problems of 1) how the model decides _which_ of the many
perceptions that preceded the reward should be repeated 2) how the model
knows how to produce (using the appropriate actions) a perception it has
never produced before and 3) how the model can repeat an uncontrolled
perception is _has_ produced before when circumstances demand a new set of
actions in order to producethat perception, the model's main problem is that
it only learns to repeat the specific perceptions (acts) that have been
followed by reward. For example, the model will only repeat the _specific_
movement perceptions that produced reward on the previous occasion. If the
movements required to produce reward are different on the next occasion, the
model will be completly flummoxed until it manages to produce the movement
perceptions that happen to produce the reward.

The problem with your dark side model (as described) is that it learns to
repeat acts (perceptions that occur before the reward). Control systems can't
control by repeating acts; they control by varying the references for lower
order acts (perceptions) _as necessary_ in order to control higher order acts
(perceptions). Control systems control by using the appropriate _transfer
function_ relating error to output (even if the output is a reference
specification for the perception controlled by a lower order system). So, in
order to learn to control new perceptions (or old perceptions in new
environments), a control system must learn _transfer functions_.

One more point. You say:

I'm hypothesizing that it [the rat] remembers what it was doing when the
pellet appeared and uses the remembered reference specifications for those
acts to repeat those acts. That severely limits the number of perceptions
that need to be repeated.

This is more dark side thinking. When you look at a rat, all you see is one
or two activities that produce reward. But that's YOUR perception of the
situation -- actually, it's just the aspect of your perception of the
situation to which you attend. The rat is actually producing (and
controlling) many, many "acts" (perceptions) "when the pellet appears". A rat
(like everyone else) is always controlling all the perceptions it controls;
it's controlling the force at its joints, its center of gravity, it's
movement patterns (transitions), it's relationship to other objects, etc,
etc. The rat doesn't stop controlling all these variabls and just control
"the lever" at some point. Pressing the lever is just one visible side
effect of the rats continuous control of a zillion perceptual variables
that it is continuously controlling. The lever press is the side effect you
notice, however, because you know that that is the one event that will
produce reward. But the rat is just busy controlling everything it controls -
- its saliva flow, its surface temperature, its balance, its gait, its
movements etc etc - - _while_ it is controlling its orientation relative
to something in the cage (lever), the visual patterns on its retina, the
auditory patterns, etc etc.

When you write your response to this post, try to think about all the
perceptions you are controlling _besides_ your perception of your reply to
this post. Your attention may be focused on your reply but try to move your
attention to other aspects of your perceptual experience - - your posture,
the music in the background, the smell of turkey cooking (they do eat turkey
on the dark side, don't they?), etc etc. All these perceptual variables are
there, and most are under control, even when you think that the _only_ thing
your are controlling (your only "act") is your response to the post.

It is this rich, personal world of perceptual experience -- a world that
demonstrably (using PCT methods) exists in other organisms --that is not seen
by those on the dark side. This is why those on the dark side are willing to
use "behavior modification" for "the good" of another organism. This is why
I work with Luke to throw light on the dark side; and this is why PCT is the
dark side's worst nightmare;-)

Best

Han Solo

[From Bruce Abbott (961127.1245 EST)]

Rick Marken (961127.0840) --

This arrives as I'm heading off to class, but in a quick scan I noticed
something missing in your reply.

You didn't answer any of my questions.

An oversight?

Regards,

Bruce

[From Bruce Abbott (961127.2010 EST)]

Rick Marken (961127.0840) --

But let me quickly explain why your learning model will not learn to control.
The basic problem (among MANY) is that your model, as described, simply
causes a reference to be set for the _specific perception_ (act) that
occurred prior to the reward.

I follow you so far . . .

Ignoring the problems of 1) how the model decides _which_ of the many
perceptions that preceded the reward should be repeated 2) how the model
knows how to produce (using the appropriate actions) a perception it has
never produced before and 3) how the model can repeat an uncontrolled
perception is _has_ produced before when circumstances demand a new set of
actions in order to producethat perception, the model's main problem is that
it only learns to repeat the specific perceptions (acts) that have been
followed by reward. For example, the model will only repeat the _specific_
movement perceptions that produced reward on the previous occasion. If the
movements required to produce reward are different on the next occasion, the
model will be completly flummoxed until it manages to produce the movement
perceptions that happen to produce the reward.

I predict that the _rat_ will be completely flummoxed (for a time) if the
requirement for producing the reward are suddenly changed, unless the the
new requirement is likely to be met in the course of attempting to reproduce
the old act. Not only that, I predict that the standard (nonlearning) PCT
model would be completely flummoxed by the same change.

I do have in the back of my mind the idea that the failure of the old act to
"produce" would lead to the repetition of previous acts that were successful
in the past (and now being recalled from memory), but I don't have much of
an idea yet how to construct the mechanism that might have this result, just
a short list of its requirements.

The problem with your dark side model (as described) is that it learns to
repeat acts (perceptions that occur before the reward). Control systems can't
control by repeating acts; they control by varying the references for lower
order acts (perceptions) _as necessary_ in order to control higher order acts
(perceptions).

You start by asserting that "control systems can't control by repeating
acts," but then show how they can control _to_ repeat acts, as if the truth
of the first statement somehow depends on the falsity of the second. But
this is nonsense: both are true. By controlling _to_ repeat a given act, a
rat, pigeon, or person can gain control over yet another perception. For
example, by producing a lever press (a controlled act-perception), the rat
gains control over food delivery.

Control systems control by using the appropriate _transfer
function_ relating error to output (even if the output is a reference
specification for the perception controlled by a lower order system). So, in
order to learn to control new perceptions (or old perceptions in new
environments), a control system must learn _transfer functions_.

The rat must only learn that, to produce food, it must first produce a
certain set of acts (i.e., approach and depress the lever). As the rat
already knows how to control its own movements (i.e., it has already learned
the relevant transfer functions), it need only learn which controlled
movements to produce. This means identifying which movements are followed
by pellet delivery. Because the rat already knows how to make any movements
it wants to, all it needs in order to obtain food is some reason to want to
approach and press the lever -- it need not learn any new transfer
functions. As the movements it was executing at the time the pellet was
delivered are normally the most likely to have produced the food pellet (if
indeed the movements were involved at all), the rat's best bet is to repeat
those movements, as best it can remember them. Whatever mechanism normally
selects which among numerous possible acts to perform (and you seem vague
about this mechanism yourself), it seems reasonable to assume that the
appearance of the food (which thus removes the error in the food-seeking
system) would lead to the playback of the memories of the just-completed
acts, and that organisms such as the rat have the ability to control the
references necessary to recreate those remembered acts.

How is this done? I don't know. And according to your answer, neither do
you. This, it seems to me, is an interesting and important problem for HPCT
research.

One more point. You say:

I'm hypothesizing that it [the rat] remembers what it was doing when the
pellet appeared and uses the remembered reference specifications for those
acts to repeat those acts. That severely limits the number of perceptions
that need to be repeated.

This is more dark side thinking. When you look at a rat, all you see is one
or two activities that produce reward. But that's YOUR perception of the
situation -- actually, it's just the aspect of your perception of the
situation to which you attend. The rat is actually producing (and
controlling) many, many "acts" (perceptions) "when the pellet appears". A rat
(like everyone else) is always controlling all the perceptions it controls;
it's controlling the force at its joints, its center of gravity, it's
movement patterns (transitions), it's relationship to other objects, etc,
etc. The rat doesn't stop controlling all these variabls and just control
"the lever" at some point. Pressing the lever is just one visible side
effect of the rats continuous control of a zillion perceptual variables
that it is continuously controlling. The lever press is the side effect you
notice, however, because you know that that is the one event that will
produce reward. But the rat is just busy controlling everything it controls -
- its saliva flow, its surface temperature, its balance, its gait, its
movements etc etc - - _while_ it is controlling its orientation relative
to something in the cage (lever), the visual patterns on its retina, the
auditory patterns, etc etc.

That's the problem for rat or human: how to identify which of these possible
changes produced by its own movements may have triggered the release of the
pellet. This is why I suggested that there must be some kind of integration
of experiences going on, so that the movements reliably correlated with
pellet delivery can be separated from those only incidentally related (a
process to increase the signal-to-noise ratio). It's a serious problem, but
not insoluble. But the only perceptions needed for this analysis in order
to identify the correct _acts_ to perform are act-perceptions: those
resulting from the organism's own movements.

When you write your response to this post, try to think about all the
perceptions you are controlling _besides_ your perception of your reply to
this post. Your attention may be focused on your reply but try to move your
attention to other aspects of your perceptual experience - - your posture,
the music in the background, the smell of turkey cooking (they do eat turkey
on the dark side, don't they?), etc etc. All these perceptual variables are
there, and most are under control, even when you think that the _only_ thing
your are controlling (your only "act") is your response to the post.

Unless I just _changed_ my posture, it is not relevant, not an act. Nor is
the music in the background, nor the smell of turkey cooking. (I _can_
learn that the smell of turkey cooking means that I will probably find a
turkey in the oven, and so on, but that is another, although related,
story.) The number of perceptual variables changing as a result of my own
movements is rather limited. For example, if a doorway is low and I fail to
notice, I may hit my head on the top of the door frame. The number of
movement-induced perceptual changes that immediately precede the sharp pain
arising from my scalp is small; I have little difficulty identifing what I
was doing at the time my head struck the frame.

You seem intent on trying to demonstrate that my little half-developed
proposal could not possibly work, even with further development. I do
appreciate your criticisms, as they help to shape my thinking. But why not
also try to think the situation through with me to find a way for it to work
and remain consistent with "basic PCT"? As they say, two heads are better
than one.

This is why
I work with Luke to throw light on the dark side; and this is why PCT is the
dark side's worst nightmare;-)

I think with regard to this problem that we are all somewhat in the dark,
PCT or no PCT.

Regards,

Bruce

From Tracy Harms (1996;11,27.19:12 MST)

Bruce Abbott (961127.2010 EST)]

Rick Marken (961127.0840)[...]
When you write your response to this post, try to think about all the
perceptions you are controlling _besides_ your perception of your reply to
this post. Your attention may be focused on your reply but try to move your
attention to other aspects of your perceptual experience - - your posture,
the music in the background, the smell of turkey cooking (they do eat turkey
on the dark side, don't they?), etc etc. All these perceptual variables are
there, and most are under control, even when you think that the _only_ thing
your are controlling (your only "act") is your response to the post.

Unless I just _changed_ my posture, it is not relevant, not an act. Nor is
the music in the background, nor the smell of turkey cooking.

This looks very wrong to me. There are perceptions being actively
controlled by a given posture even when that posture goes unchanged. Every
posture is the result of a combination of muscular tensions, and the nature
of muscles means that there are error-corrections which alter these muscle
tensions over time as a given posture is held *unchanged*. Stand on one
foot, as still as you are able, for five, twenty, or fourty minutes; that
should make this clear. Meditation and military training are means by
which people become skillful at maintaining given postures for long
durations with little or no apparent change, but a small amount of personal
practice quickly teaches that prolonged maintenance of a given posture is
accomplished by constant error-correcting behavior--just as PCT theorizes.

Hearing ambiant music and smelling ambiant odors are also relevant. I
haven't figured out how to rightly think about these things yet, but I have
figured out enough to be confident that ongoing perceptual-control
correction is involved in both. Sorry I'm too much the tyro to spell out
the how and why.

Tracy Bruce Harms tbh@tesser.com
Boulder, Colorado

···

==========================================================================
   "There is only one way I can see for fallible, ignorant human beings
   to live in accord with their own real natures and that is to discard
   forever the principle of controlling each other's behavior, dropping
   even the *desire* to control other people, and seeing at every level
   the fallacy in the logic that leads to such a desire."
                                                   William T. Powers

[From Bruce Abbott (961128.1120 EST)]

Tracy Harms (1996;11,27.19:12 MST) --

Bruce Abbott (961127.2010 EST)

Unless I just _changed_ my posture, it is not relevant, not an act. Nor is
the music in the background, nor the smell of turkey cooking.

This looks very wrong to me. There are perceptions being actively
controlled by a given posture even when that posture goes unchanged. Every
posture is the result of a combination of muscular tensions, and the nature
of muscles means that there are error-corrections which alter these muscle
tensions over time as a given posture is held *unchanged*. Stand on one
foot, as still as you are able, for five, twenty, or fourty minutes; that
should make this clear. Meditation and military training are means by
which people become skillful at maintaining given postures for long
durations with little or no apparent change, but a small amount of personal
practice quickly teaches that prolonged maintenance of a given posture is
accomplished by constant error-correcting behavior--just as PCT theorizes.

I agree completely that we are actively controlling perceptions all the
time, just as you (and PCT) describe. But it is the consequences of
movements that I want the model to focus on -- the dynamic changes in the
values of these movement-related perceptions. At least, that seems to make
sense to me. I'm open to suggestions.

Hearing ambiant music and smelling ambiant odors are also relevant. I
haven't figured out how to rightly think about these things yet, but I have
figured out enough to be confident that ongoing perceptual-control
correction is involved in both. Sorry I'm too much the tyro to spell out
the how and why.

Relevant to what? Let's say that you're trying to figure out how to get the
refridgerator door open. What do the music and ambiant odors have to do
with that problem?

  "There is only one way I can see for fallible, ignorant human beings
  to live in accord with their own real natures and that is to discard
  forever the principle of controlling each other's behavior, dropping
  even the *desire* to control other people, and seeing at every level
  the fallacy in the logic that leads to such a desire."
                                                  William T. Powers

So if my neighbor wants to kill me, I should let him do it, right?

Regards,

Bruce

From Tracy Harms (1996;11,28.10 MST)

Bruce Abbott (961128.1120 EST), in response to Bill's sentence which
appears in my .sig you quip

So if my neighbor wants to kill me, I should let him do it, right?

I am pretty much stunned by the facetious shallowness of your reply. This
is the sort of conversation I brace myself for when I go to UseNet's
alt.politics.whatever, but not at all what I expected here. In my
judgement this is not a genuine question, but an expression of
belligerence. Frankly, Bruce, I'm not in the mood for an argument of that
sort.

Tracy Bruce Harms tbh@tesser.com
Boulder, Colorado

···

==========================================================================
   "There is only one way I can see for fallible, ignorant human beings
   to live in accord with their own real natures and that is to discard
   forever the principle of controlling each other's behavior, dropping
   even the *desire* to control other people, and seeing at every level
   the fallacy in the logic that leads to such a desire."
                                                   William T. Powers

[From Rick Marken (961129.1420)]

Bruce Abbott (961127.1840 EST) --

These seem to be serious gaps in what was supposed to be an HPCT-based
explanation, if I understood you correctly...I dare say that it hardly seems
less speculative than the little story I offered . . . Or did I miss
something?

The speculativeness of your story wasn't the problem; it was the
plausibility of your story, given what we know about how behavior
(control) works.

What empirical observations DO you see as offering difficulties for HPCT
as now constructed? Surely after 14 or so years you've run into _some_.

Are you asking "what empirical observations _have already been_ made that
offer difficulties for HPCT as now constructed?" If so, then obviously the
answer is "none". Otherwise we would already have fixed the model or
abandoned it.

The closest I ever came to making an observation that "offered difficulties"
for HPCT was when I discovered the "Marken effect"; the fact that control
is better when the disturbance to a controlled variable is the output of a low
gain control system rather than an inanimate source generating the _same_
disturbance waveform. A simple control model exhibits identical behavior
in both disturbance conditions but people do much better against the "live"
disturbance than against a playback of the _same_ disturbance waveform.

This result came as a surprise to both Bill P. and me. Bill quickly found the
solution, however. The control model acts just like the human when a transport
lag is included in the model. So a result that appeared at first glance to
offer difficulties for the HPCT model actually provided more support for it.

Bruce Abbott (961127.2010 EST)

You start by asserting that "control systems can't control by repeating
acts," but then show how they can control _to_ repeat acts.

Your problem here is that you are ignoring the hierarchical nature of
control. In a hierarchical control organization, a control system at
any level above the lowest in the hierarchy must vary its _actions_ (the
reference for lower order "acts") in order to produce an "act" ( the
intended state of its own controlled perceptual variable). Your model says
that a control system repeats a particular action (the reference for a lower
order "act", such as pressing a lever) when a reward occurs. Again,
I suggest that you build the model and see what happens. I don't believe
that such a model could possibly learn to control.

The rest of your post was nicely answered by Tracy Harms(1996;11,27.19:12 MST)
and Bill Powers (961128.2200 MST).

Tracy Harms (1996;11,28.10 MST) --

Bruce Abbott (961128.1120 EST), in response to Bill's sentence which
appears in my .sig you quip

So if my neighbor wants to kill me, I should let him do it, right?

...Frankly, Bruce, I'm not in the mood for an argument of that sort.

I think argument can be useful, especially on this important topic. Bruce
Abbott was responding to the following quote from B:CP:

"There is only one way I can see for fallible, ignorant human beings
to live in accord with their own real natures and that is to discard forever
the principle of controlling each other's behavior, dropping even the

>*desire* to control other people, and seeing at every level the fallacy

in the logic that leads to such a desire."

I think this is a brilliant choice of quotes for your sig file, Tracy. It
really captures the "bottom line" of PCT. Of course, in order to understand
the bottom line,it's necessary to understand what was above it. Apparently,
you do but Bruce Abbott is still having some problems with it.

Bruce seems to think that "discarding forever the principle of controlling
each other's behavior" is an admonition to "stop all controlling" or, in other
words, to "be a door mat". Of course, we can't stop controlling, whether
we want to or not. But we can stop some of our controlling -- the controlling
that involves control of other living control systems.

If you don't want your neighbor to kill you, then there are basically
three ways to deal with the problem; 1) control him 2) negotiate with
him or 3) defend against him. PCT suggests that option 1) always
leads to problems -- ergo, Bill's statement above. Option 3) is what you
do, whether you want to or not, when all else fails (or is impossible); if
you are controlling for not getting hurt then you do what you can to resist
disturbances to this variable. Option 2) is the "best" option; PCT shows
how to negotiate effectively; the goal of negotiation is to help _both_
parties go "up a level".

Best

Rick

From Tracy Harms (1996;11,30.20)

Rick Marken (961129.1420)

Are you [Bruce Abbott] asking "what empirical observations _have already
been_ made that offer difficulties for HPCT as now constructed?" If so,
then obviously the answer is "none". Otherwise we would already have fixed
the model or abandoned it.

Perhaps things are very different in psychology than they are in
philosophy, but in philosophy serious problems may be found with a theory
yet the theory may remain popular for generations. Searches for fixes are
undertaken, but a theory's abandonment often does not widely occur until
something better comes along. (In some cases, it takes much longer
still...)

Tracy Harms