<[Bill Leach 960122.13:03 U.S. Eastern Time Zone]
Remi Cote 012196.1050 Montreal Quebec
In ethology, Lorenz describe an if-then model that fit with
observation. If there is a drive, and the right stimulus he
postulated a release mechanism that trigger a precise pattern
of action, and these action are in turn stimulus and trigger other
action in other individal who are involve aggressive or sexual ritual.
In PCT, the problem with this sort of explanation is that it is
explicitely NOT a description of how a control system functions.
To try to explain the difference let me use your example:
Start with room temperature at 25 degrees and the thermostat setpoint at
20 degrees (as you used in your example).
If you apply a "stimulus" of dropping the room temperature by 6 degrees
(some how -- though in Montreal right now you probably could get close to
a "step change" by just opening the window <grin>) the observed behaviour
of our example temperature system will be to "emit" an output action
(turn on the heater) _as though_ the temperature system was "triggered"
into action by our stimulus.
What is wrong with this sort of description/explanation? In the first
place the drop in temperature does not _independently_ result in the
system's action. Our initial assumption that the thermost setpoint was
at 20 degrees is _equally_ important.
If we explain what we observed with the behaviour of the temperature
system by saying that our action of dropping the temperature by greater
than 5 degrees (or even that if we drop the temperature to less than 20
degrees) "stimulate" the system into action we then have an explanation
that is useless for explaining why when the setpoint is raised six
degrees (with the same initial conditions) our example system "emits"
behaviour without external "stimulus".
Such an explanation as Lorenz's may well be an exact match to what has
been and can be observed but it _only_ will be such a match _when_ the
internal independent variables are the same as they were for his
experiments.
He is failing to recognize a "control process" and apply control system
theory based analysis to a control process (or at least even if he does
recognize the existance of a control process he is not using control
process analysis which by definition can not be correct if it ignores the
reference signal).
We have an animal painter in Quebec, gisele Benoit, who work with moose.
She use a fainted panache to deceive male moose. With the panache, she
can elicit at will aggressive pattern of behavior and happily for her
(she's been charge only once) as soon she dropped the fainted panache
the animal calm down.
Having lived in Idaho some years ago, I have a bit of experience with
Moose (more with Elk -- which are also rather impressive animals). Few
people that have not had close contact with something like horses have
any real concept of just how powerful such creatures are!
Your description is again a "Stimulus-Response" description. A control
systems description would assert something along the lines:
When your friend displays the fake panache and the moose perceives it,
a controlled perception is disturbed and control system action occurs in
an attempt to restore the perception to its' controlled value.
As a practical matter this might prove to be a bit difficult with a
1,000KG moose but the control theorist would then attempt to set up an
experiment to determine what perception was disturbed (ie: What
perception is the moose controlling when the panache is displayed?).
In actual research efforts, it is here that PCTer differ most sharply
from other researchers. The both the non-control theorist and the
control theorist begin conjecture about why the moose charges but the
control theorist is doing so to enable design of specific experiments.
The "conventional" researcher is typically only looking for a "rational"
or "plausible" explanation -- an explanation that typically may never be
tested.
As a result, much of the behaviour of animals is attributed to "reasons"
that the animal would probably view as ludacrous if it were able to think
at such a level at all.
Other ethologist have work in other domain with human baby... John
Archer (1992) (ethology and human development) describe attachment
theory of Bowlby as a "behavioral system" in wich there is a top-level
set goal: representation of proximity, and If the discrepency signal
received by the baby are high there is a non-goal directed behavior
(crying), discrepancy medium a goal directed behavior (calling), small
discrepancy: subgoal (looking, following). And if there is a perception
of a decrease of discrepancy, the repertoire change.
This is a bit "better" from a control theory perspective (until the
"output" is discussed).
From a PCT standpoint the term "non-goal directed behaviour" is
non-sequitar. The output is always goal _directed_ even if it is
completely ineffective. The researcher/observer calling behaviour
"non-goal directed" is only admitting to not understanding what the
observed organism is doing. A problem, of course, is that when an
oganism does not have all of the necessary components of a control loop
(or system) to achieve successful control of an important perception the
resulting "random" behaviour can not be analyzed (at least currently) in
any precise and repeatable manner.
A serious problem (from a research point of view) is that reorganization
makes "one time changes" to the structure of the control loops and such
changes are (at least) somewhat random. This with the individual
subject, there is no way to "repeat" the experiment since you are not
dealing with the same structure after a reorganization as you were before
it occurred.
With multiple subjects the problem is that reorganization is a random
process and even if observations seem to show otherwise, a change in one
individual is not exactly the same as any other.
These are not exactly insurmountable problems but they definately add
difficult to deal with complexity to any control systems type studies of
learning.
In PCT we avoid actually trying to make specific claims as to what
perceptions might be "ultimate" references (or goals if you will). What
in PCT we call "intrinsic references". Thus, while I think most of us
would agree that somehow we all (that is universally -- no exceptions)
has some sort of reference related to perceiving a consistent or reliable
"world"... in some way, I don't believe that anyone here would try to
"pin it down" to a specific and exacting definition.
Though far from being proof that "a reliable world" is somehow possibly
closely related to an "intrinsic reference" virtually every animal (at
least) has been observed to "react strongly" and usually "suddenly" when
a "normal experience" fails to proceed "normally". Personally we usually
employ the term "surprise" when describing are own experiences with such
an event.
Much of "PCTers problem" with other research is not even necessarily that
the other researches are failing to accurately describe what was observed
but rather that it appears that other researcher either do recognize
control phenomenon or if they do recognize control phenomenon the do not
understand it well enough to know what observation are important and what
observations are not important.
This even researchers that talk about control usually do so in a manner
that makes it quite clear to control theorists that the researcher do not
even have a "clue" as to how a control system functions nor what makes up
a control process.
Very well known researchers display a complete ignorance of the
significance of even the slightest movement by a living organism with
their "almost off-hand" use of such expressions as "the organism emits
behaviour". Other than for a "closed loop negative feedback control
system", reliable and repeatable actions are just plain not possible in
the "real world" with all of its' variation and disturbance.
-bill