DA's problem;Fred&ginger;Martin's math; fury

[From Bill Powers (930415.0745 MDT)]

Bill Cunningham (930414.1200)

The problems you describe are horrendous indeed. Perhaps the real
answer is to be found not in solving the problem, but in changing
the problem.

I can cite some horror stories, but 1 analyst trying to wrestle
with 6000 disparate data bases may give you some sense of the
overload problem.

The main difficulty here, it seems to me, lies in the term "data
base." The mere fact that something has been written down does
not make it data. If one analyst is handling 6000 "data" bases,
what comes out is bound to be mainly the product of imagination,
free association, and personal prejudice. This sounds to me like
a problem that has to be handled where the so-called data goes
in, not where it comes out.

... the conditional branching associated with programming
really involves choosing alternate courses of action.

In PCT, what comes out is not a course of action, but a
specification for perceptions. If we take the proposed levels
seriously for a moment, what comes out of the program level is a
set of specifications for sequences (i.e., perceived orderings
without branch-points: lists). No sequence of actions is
necessarily implied; the actions required to produce a certain
sequence of results are not known, in general, because disturbing
factors in the environment are not, in general, known in advance.
It is up to the lower-level systems to determine at every moment
what must actually be done by way of output to keep the specified
sequence going in perception. If I tell you to (1) line up the
kindergarten children, (2) march them down the hall, and (3) take
them outdoors for a fire drill, there is no set of predefined
actions that will result in the occurrance of this sequence. You
simply do what it takes to perceive the elements of this sequence
occurring in the specified order ("Johnny, get back in line.
Where is Linda? Stay away from that water cooler. Linda! No,
children, I said no water, we have to get in line now..."). PCT
shows the futility of planning outputs, as does real-life
experience.

So the poor DA has wade through a surplus of data, reject the
irrelevant, compress it to a level of abstraction that is
meaningful in terms of the decisions to be made, and then deal
with the uncertainty.

This is why I say that the product is most likely to reflect the
characteristics of the DA. The worse the data, the more likely
one is to come up with spurious conclusions from it and think
they are meaningful -- especially when there is pressure to
arrive at a conclusion.

The function of information is to remove/reduce the uncertainty
at each branch of the tree.

This is the pressure that leads people to read into the source
material anything that sounds halfway plausible. Information
can't read things into the data. Uncertainty is subjective,
depending on what one decides to imagine as well as what one
perceives. I am quite convinced of the utility of statistical
procedures in reducing inner conflict about what to do. I am
unconvinced that such conflicts ought to be removed regardless of
the quality of the data, or that bad data can lead to choices
that are in fact the best possible.

The hierarchy of abstraction provides tremendous data
compression, and the higher order references provide the basis
for filtering.

But they also provide the basis for reading into any situation
what your own prejudices lead you to expect to find. Bad enough
data and a complex enough way of handling it can provide the
excuse for acting just as you thought you should act, anyway.
During the Cold War, no amount of data could have convinced a
hawk that the Soviet Union was not quivering on the brink of an
all-out attack on the USA, and with bad enough data, an analyst
sympathetic to that view would be able to extract indicators
showing that exactly this would happen, but for our overwhelming
war machine. Of course an analyist sympathetic to a different
view could use exactly the same data to arrive at predictions
supportive of the doveish view. When you start with bad enough
data, the decisive factor is not the method of analysis, but
human imagination and belief.

ยทยทยท

-------------------------------------------------------------
Greg Williams (930414 - 2) --

Thanks for the comments. Looking forward to the next installment
on Maier. Yes, the comments on his work sound familiar!

By the way, I DO appreciate the fact that he reports what ALL his
rats did, and considers each type of result worthy of discussion.
-------------------------------------------------------------
Dan Miller (930414.1315) --

... it seems to me, and correct me if I am wrong, that their
(Fred and Ginger's) social behavior just seems to me to be more
than observable side effects.

How can you be wrong about the way things seem to you?

What can the observed behavior of Fred and Ginger be but a side-
effect, for an observer who can't see the world from Fred or
Ginger's point of view, feel what each feels the other doing,
feel the efforts and see the relationship of the other to
oneself? The kinesthetic aspects of what Fred and Ginger are
doing are totally unknown to anyone else. The symmetry that the
audience sees as Fred and Ginger spin around each other and move
in the patterns of the dance is not what either Fred or Ginger
sees. Just imagine a video camera on Ginger's shoulder, and what
the screen would show. That would be nothing like what we see on
the movie screen. The mastery of Fred and Ginger lay in their
ability to imagine how their own worlds would appear to someone
else.

If you see me pick up a glass of water and drink from it, we can
agree that we have both experienced the same happening. But that
agreement is superficial. You did not feel the roundness of the
glass or its temperature. When I tip the glass to drink, I see
its bottom through the water; you see the other side of its
bottom, through air. You see the water level moving downward from
one sip to the next; I see it receding. I taste the water, feel
its wetness, sense the effort of my swallowing muscles; you can
only imagine those things. I feel my thirst being assuaged; you
are not thirsty and would not find an arbitrary drink of water at
that moment satisfying.

The social world is always divided into two parts: everyone else,
and me. What I know of everyone else consists only of what I can
see their arms and legs and torsoes and faces doing, what I can
hear their voices doing -- what I can notice from my point of
view. When others contact me, I feel the shove, but not the push.
I am always seeing my own behavior from inside, and the behavior
of everyone else from outside. All this is true of every single
"I" that exists.

The old-fashioned social sciences focussed strictly on the
outsideness of people: what we could see their exterior surfaces
doing, and doing to the world between them. The observer and the
observer's insideness were largely ignored; they played a part
only through unrecognized influences of the observer's
perceptions and goals on the observations. Only by ignoring his
or her insideness could the observer indulge in the illusion of
objectivity: just reporting the facts, ma'am.

Just consider one class of mass behavior that Clark McPhail has
noted: everyone in a stadium turns to look at a common phenomenon
like an exploding scoreboard. Even putting it that way is to
confound the outside with the inside. We do not know what these
people are looking at: we aren't privy to the retinal images of
each person, or to where in these images the attention is turned
at any moment. Even to say they looked at something is to
attribute a purpose to the turning of all those heads. Even
without knowing it, the observer is putting himself in the place
of the other, and saying "If I turned my head that way under
those circumstances, I would be intending to perceive the thing
toward which the head is oriented. I would be centering the image
of that exploding scoreboard in my visual field, and attending to
it."

If we actually described social behavior strictly in terms of
what we could observe, it would cease to make sense. You couldn't
even say that you saw me put the empty glass down on the table.
All you could say is that you saw my hand and arm descend until
the glass contacted the table, at which point my fingers opened
up and moved away, leaving the glass in a fixed position where it
was prevented from further descent by the presence of the table.
You can't see that I am doing something called "putting" or that
its intended result is to place the glass in a particular
relationship to the table. The very language in which we
naturally describe the behavior of others contains attributions
of purpose or intention, of accomplishment of some particular
result as opposed to a different result. The language represents
the inside view as much as the outside view: agency as much as
outcome. But we use it as if it were only the outside view being
described.

PCT approaches social science in a way that clearly separates the
inside and outside views (Phil Runkel wrote a book called "Inside
and Outside," from which I am undoubtedly cribbing). It shows us
that to understand the organization of anyone -- meaning
ourselves, or any one of the others, or all of the others -- we
must understand the inside view. We must understand that our
outside view of others is really an inside view from within
ourselves, the observers. We must understand that all we know of
social conventions is what we, ourselves, alone in here,
understand of them. We must understand that this is true of every
single one of the others, too. Social interactions are ALWAYS
seen from an individual's point of view; there is no other kind
of viewpoint available. Even to speak of a viewpoint is to speak
of one solitary individual -- from the inside.

All the old sciences of behavior have attempted to ignore the
presence of the observer. But even in the attempt, they have
failed to eliminate the properties of the observer. Most
behaviors are named not by naming the movements that bring them
about, but by naming the regular outcomes of those movements.
Even without the benefit of PCT, scientists have worked,
unknowingly, under the principle that outcomes, not acts, define
behavior. From PCT we understand that those outcomes could not
have been regular unless they had been intended by the actor,
unless they were under control by the actor. But from the
traditional view, it just seemed that the outcomes were the
behavior of the organism, emitted just as a muscle emits force.

No behavior -- no control.

But what does "behavior" mean? Does it mean the motor actions
that move the limbs in ways that others can see? Or does it mean
the outcomes that are perceived and maintained in particular
states through variations in the acts? If you mean the actual
outputs, the movements, then you can't deduce control just from
knowing them. Control can be seen only in the outcome, where
actions join with disturbances to yield an otherwise inexplicably
constant result. If you just look at the actions performed by
people, you will never see control. You will see the effects,
mainly, of random disturbances. When you understand PCT, you know
that the movements performed by others tell you almost nothing
about what they are controlling for; the movements indicate
mainly that disturbances are present, disturbances which you may
not have seen, or seeing them may not have seen as disturbing
some outcome of the action. When you fully understand the
actions, you will see them as mirroring the sum of all
disturbances, and only in part serving to create a particular
outcome. They tell you very little, by themselves, about what the
person is doing.

This is precisely why even the PCT-naive name behaviors in terms
of outcomes, not actions. When Fred and Ginger join hands and
whirl around a common center, you are seeing an outcome of their
actions, not their actions. And of course you are seeing this
from your own point of view, in which the actors whirl around on
a stationary stage. If you were either actor, you would see your
partner stationary in front of you, with the stage and the
audience whirling around you. You would be conscious of the
centrifugal force pulling you apart from your partner, and would
feel the action that is preventing separation -- not a whirling
action, but a direct steady pull. It is only this direct steady
pull that is maintaining the circular motion and keeping it from
turning into two independent trajectories. None of that is
evident from where you sit.

It isn't output that controls input. It's variations in output,
based on deviations of input from a reference signal, that
produces the overall effect called control. The title "Behavior:
the control of perception" does not say that outputs control
inputs. Behavior is not output. It says, when you emerge through
the back cover, that behavior is the process by which organisms
control the perceptions that matter to them. This process
involves the entire control loop, not just one part of it.

It seems like almost everything that I am interested in has
this social interaction quality to it - dancing, conversing,
making love, bargaining, negotiating, gang activities, and so
on. I know that these are only words, but aren't they all?

The words point to phenomena, and a phenomenon can be understood
if you ask how each party to the interaction knows about it, what
each party wants from it, how each party disturbs the perceptions
of the other, and how the actions of each party prevent, or
attempt to prevent, disturbances of controlled variables. Find
the controlled variables for each party to the interaction, and
you will understand the interaction.

Bill Powers tells me that we need to get past (below?) the
words to the process. I couldn't agree more. That is what we
need to do, get to the process.

Yes. We keep trying.
-------------------------------------------------------------
Martin Taylor (930415.2310) --

<Bill's demonstration with the multiplier may have come out of

his head, but I did suggest it to him.

Yes you did, but I've been building models with variable feedback
gain for a long time. A model of operant conditioning, for
example, shows the effect of varying the proportionality factor
between presses per minute of output and reinforcements per
minute of input.

I provided a new relationship, which neither Bill nor Rick
cared to comment on when saying IT can say nothing new about
PCT. There's a lot of imagination in the perceptual
realtionships involved in reading.

Well, you probably haven't been through all the outstanding posts
yet. I think there's a serious problem with your derivation of a
relationship between bandwidth and information rates -- for one
thing, the equations say that for the continuous-variable case,
the maximum effective loop gain is either 0 or 1 (I forget which,
but it's in a post to you). In all cases where (as you assert)
the bandwidth of the disturbance is greater than that of the
input function, the maximum effective loop gain is less than 1.
This is the kind of absurd result that comes out of doing
mathematical manipulations blindly. In one post I have given you
my reasons for saying that the effective bandwidth of the
disturbance is always identical to the bandwidth of the
perceptual signal, and equal to or smaller than the bandwidth of
the input function. I disagree with your premises, so how can you
expect me to suitably acknowledge and defer to the conclusions
you drew from them?
---------------------------------------------------------------
Avery Andrews (930411?.1710)--

>>I think that a bit of fury is excusable.

But probably not very effective..

I don't know. First you have to get the mule's attention.

There's also some interesting philosophical points - F & T seem
to have some kind of objection to `clever machinery' (like
hierarchies of ECS), on the basis that there's some kind of
homuncular regression involved (if you explain clever behavior
in terms of clever machinery, who designed the machinery ...),
failing to notice that things like insect bodies provide
innumerable instances of `clever machinery' arising by natural
selection.

Have you ever known anyone in a philosophical argument to goggle
at you and say "My gosh, you're right, you've completely
destroyed my argument?" It's the vagueness of ideas like "clever
machinery" that allow one to go on defending a philosophical
notion indefinitely. There's always one more way to get around
the objections. Nobody ever actually has to LOSE a philosophical
argument. All that really happens is that one side gangs up on
another side and declares that nobody believes that any more.
That may cow the losers into admitting defeat (for social
reasons), but I'm sure the losers still lie awake at night
playing I-shoulda-said.

But maybe F & T are the best target.

Where do you think we would find a journal with referees who
would not take such an attack as prima facie evidence that the
authors are ignorant of the latest developments in the field?
---------------------------------------------------------------
Best to all,

Bill P.