irrelevant side-effects

[From Bill Powers (951017.1000 MDT)]

Bruce Abbott (951016.1935 EST) --
RE: irrelevant side effects

     Perhaps it's just a semantic thing, but I just don't view phenomena
     like a child's anguish at being separated from its mother as
     irrelevant, even though it may arise from the actions of a control
     system whose perceptual input is being disturbed.

A low blow, as if those who consider observations of the visible actions
of a control system to be irrelevant side-effects don't care about a
child's welfare. The side-effects are side-effects simply because they
aren't the child's own perceptions; they are the observer's perceptions.
When the child weeps, clings to its mother, greets further loss with
pitiable stoicism, and turns its back on friendly overtures, what we are
seeing is not the problem, but only the actions that are intended to
deal with the problem. The actions are the irrelevant side-effects; they
are irrelevant because what is relevant is the set of perceptions that
they are supposed to control.

We do not observe the child's anguish. We can only imagine it, from our
own experiences, by imagining what we might be feeling if we were
producing the same actions the child is producing. But when you try to
judge what the problem is in this way, you can't recognize it if the
actions don't suggest anguish to you -- if the child is using a means of
control that you have never used. Also, you might conclude that if you
can get the child to stop crying, stop clinging to its mother, stop
rejecting friendship, stop presenting a stony face to the world, you
have then solved its problem. Instead, you might well have taught the
child that the only way to keep you from adding to its anguish is for it
to avoid showing the symptoms that get you on its case.

PCT says that we should consider overt actions only insofar as they can
help us to identify controlled variables. The action in itself is
irrelevant; if circumstances were different, so that a particular action
would not work against disturbances of the controlled variable, a
different action would be used -- but the same problem would exist.

When a parent says "Don't worry, I'll be right back, here's a nice new
toy you can play with, here's some ice cream, here's a nice lady who
will play games with you, let's see a little smile...." the parent is
simply trying to make the child's actions change. This is called
treating the complaint, the symptoms. It is not treating the problem.
Perhaps if we knew more about the kinds of controlled variables and
error conditions that are involved in "anguish" we could think of
something more effective, something that would actually solve the
problem instead of simply trying to deny to the child the only present
means the child knows of to counteract the error.

The first step toward finding such solutions is to start viewing the
actions of an apparently distressed child as insignificant in
themselves, and start focusing on what they are meant to control, and
perhaps on how the child got the idea that these means are effective.

     If there is a specific action or set of actions the system
     inherently takes, this would be an important bit of information to
     know about, don't you think? What if there is a predictable series
     of adjustments that takes place during reorganization, after the
     child's tears and other responses fail to bring its mother back?
     Irrelevant? Unimportant?

Don't forget that actions are normally labeled by their outcomes, not in
terms of the actual outputs. To create a "predictable series of
adjustments" in a variable environment requires that those adjustments
be themselves controlled variables; to make the same adjustment on
different occasions generally requires different actions.

Seeing the same adjustments being made repeatedly could indicate that
there is a built-in control system that controls for a sequence of
perceptions, with a built-in reference sequence. Or, at least as likely,
it could mean that the environment is structured so that only a
particular sequence can have the intended effect, all other sequences
that have been tried having been ineffective. In the latter case, the
predictability of the sequence indicates only some predictability of the
environment, and doesn't necessarily indicate anything important about
the organism. I probably wouldn't look very hard for learned methods in
a bug, but that is what I would look for first in a human being.

The danger inherent in behaviorism is that of confusing the means of
control with the object of control -- of seeing overt behavior as an end
in itself, like seeing crying simply as an expression of an internal
state caused by traumatic stimuli rather than as a means to an end. When
we look at other people, we simply _don't see_ what they are really
doing, because without a rather extended investigation, we can't even
guess at what perceptions they are controlling. All we can see is the
set of movements, sounds, and bodily or facial configurations by means
of which they attempt to counteract disturbances of their perceptions.

···

-----------------------------------------------------------------------
Best,

Bill P.

[From Bruce Abbott (951017.2110 EST)]

Bill Powers (951017.1000 MDT) --
   Bruce Abbott (951016.1935 EST)
RE: irrelevant side effects

    Perhaps it's just a semantic thing, but I just don't view phenomena
    like a child's anguish at being separated from its mother as
    irrelevant, even though it may arise from the actions of a control
    system whose perceptual input is being disturbed.

A low blow, as if those who consider observations of the visible actions
of a control system to be irrelevant side-effects don't care about a
child's welfare. The side-effects are side-effects simply because they
aren't the child's own perceptions; they are the observer's perceptions.
When the child weeps, clings to its mother, greets further loss with
pitiable stoicism, and turns its back on friendly overtures, what we are
seeing is not the problem, but only the actions that are intended to
deal with the problem. The actions are the irrelevant side-effects; they
are irrelevant because what is relevant is the set of perceptions that
they are supposed to control.

Unleashing a "low blow," as you chose to phrase it, was not my intension.
As I think a made very clear in my post, my problem with the phrase
"irrelevant side-effect" as a way to characterize what researchers using
"conventional methods" have been studying is that it tends to suggest rather
strongly that what these researchers study is "irrelevant" in some more
general sense; that it is uninteresting, useless, a waste of time,
unimportant. My argument and example were intended to show, not that "those
who . . . don't care about a child's welfare," but that these phenemona are
far from uninteresting and that research on them is far from useless,
unimportant, "irrelevant." I attempted to show instead that the simple
observation of these "irrelevant side-effects" could yield important
insights as to the nature of the control system whose behavioral actions are
being observed in these cases. The argument has nothing whatsoever to do
with anyone's personal concern (or lack thereof) for a child's welfare.

My argument has never been that conventional methods are a satisfactory
substitute for individual-centered PCT methods (e.g., testing for the
controlled variable; constructing generative models). By responding to the
argument I do make as if I had argued in favor of this other, you have
simply muddied the waters.

    If there is a specific action or set of actions the system
    inherently takes, this would be an important bit of information to
    know about, don't you think? What if there is a predictable series
    of adjustments that takes place during reorganization, after the
    child's tears and other responses fail to bring its mother back?
    Irrelevant? Unimportant?

Don't forget that actions are normally labeled by their outcomes, not in
terms of the actual outputs. To create a "predictable series of
adjustments" in a variable environment requires that those adjustments
be themselves controlled variables; to make the same adjustment on
different occasions generally requires different actions.

I'm not forgetting that, as my later statement in the same post should have
made clear:

What I see in behavior is evidence of control, not irrelevant side-effects
of control. As an example, consider the complex courtship rituals of some
birds. What is often observed is a complex series of movements which are
manifestations of the workings of a sophisticated control system operating
at at least the "control of sequence" level, and probably higher. Although
the specific muscular contractions involved change dynamically according to
accidents of terrain, position of the partner, and so on, the general pattern
is clearly visible to the observer, stable over repetition, and follows
essentially the same form and sequence in all members of that species and
sex in which it occurs.

Seeing the same adjustments being made repeatedly could indicate that
there is a built-in control system that controls for a sequence of
perceptions, with a built-in reference sequence. Or, at least as likely,
it could mean that the environment is structured so that only a
particular sequence can have the intended effect, all other sequences
that have been tried having been ineffective. In the latter case, the
predictability of the sequence indicates only some predictability of the
environment, and doesn't necessarily indicate anything important about
the organism. I probably wouldn't look very hard for learned methods in
a bug, but that is what I would look for first in a human being.

Compare your paragraph above with mine:

The apparent complexity in the behavior is simply a reflection of the
complexity of the environment. If ten ants follow the same path on their
way back to the nest, this may reflect the fact that the disturbances in
their path produced the same pattern of counteracting actions and nothing
more, yet it might appear to the observer that the aunts' neural networks
contained some kind of inherited motor program that steered them first this
way, then that, in a reproducable, sterotypical pattern.

So, we're both saying that the environment can constrain the sequence, thus
necessitating proper tests to determine which interpretation is correct. We
agree.

The danger inherent in behaviorism is that of confusing the means of
control with the object of control -- of seeing overt behavior as an end
in itself, like seeing crying simply as an expression of an internal
state caused by traumatic stimuli rather than as a means to an end. When
we look at other people, we simply _don't see_ what they are really
doing, because without a rather extended investigation, we can't even
guess at what perceptions they are controlling. All we can see is the
set of movements, sounds, and bodily or facial configurations by means
of which they attempt to counteract disturbances of their perceptions.

Yes, of course. But I was never arguing elsewise. What I _was_ arguing is
that behavioral observations such as those collected by ethologists can
provide some important insights concerning the nature of the underlying
control mechanism, which can provide an excellent starting point for a PCT
analysis. As such they are neither unimportant, nor trivial, nor
"irrelevant" in the less narrow sense of the term. The phrase "irrelevant
side-effect," while perhaps technically true when used to characterize some
of these behavioral actions, tends to obscure this fact. Rick's apparent
belief that conventional findings are useless because they are about
"irrelevant side-effects of control" seems to me to arise from a confusion
of the broader and narrower meanings of the term.

Regards,

Bruce