[From Bill Powers (990802.0035 MDT)]
Marc Abrams says:
Isaac, I know that you feel very strongly about PCT and much less so about
HPCT ( The Hierarchy ). The main reason being that for all intents and
purposes, the Hierarchy in _any_ form has as yet been "validated" by
"scientific" evidence.
Isaac:
Yes.
Marc:
some of the lower levels have been researched but the higher ones have not.
Am I correct in this assumption?
Isaac:
Very accurate ass[ess]ment of my take.
Let's be careful about what we mean by "scientific evidence." I don't
believe that "scientific" evidence is different from any other kind: either
there is evidence or there is not.
Consider what we accept as evidence of control. If we see someone acting in
such a way as to maintain some variable in a given state despite
disturbances that, without those actions, would change the state of the
variable, and after we have ruled out illusions and coincidences, we say
that the variable is controlled by the person. This is what we in PCT have
_agreed_ to mean by control. Therefore the only "evidence" we require in
scientific terms is evidence that the action and its relationships to
disturbances did actually take place as reported. In short, all we require
is an honest report of what happened, along with the universal scientific
assumption that if any person could have seen it happen, it must have
happened in reality. That latter assumption is the weakest point of science
itself, but we manage to get by nonetheless.
Now consider what we might accept as evidence of hierarchical control.
Hierarchical control takes place when the action involved in controlling
some particular variable is itself observed to involve the control of other
variables which we call the "means" of control. If most of the means of
control are observable as control processes in themselves as in the
paragraph above, then we agree to say we are observing hierarchical
control. The only "evidence" required is evidence that an honest observer
reported relationships among variables that anyone could see, and that the
observed relationships were those of hierarchical control.
The evidence for hierarchical control is of exactly the same nature as the
evidence for control. It is a report by an honest and competent observer
that certain well-defined relationships among variables were seen. To that
we can add all the usual provisos: that the conditions of observation can
be reproduced by anyone with the requisite skill and equipment, and that
when they are reproduced the same observations are made, and other such
details that we have learned from experience are important.
This sort of evidence simply establishes the reality and replicability of a
phenomenon. It does not offer any explanation for it, but no explanation is
required if the only purpose is to establish the observational reality of
the phenomenon. I believe that in PCT we have adequate evidence for the
observational reality of both plain control and hierarchical control, and
that if anyone falls into a state of doubt about either of these phenomena,
a few minutes of systematic observation of almost _any_ behavior will
eliminate the doubt.
The particular hierarchical relationships among perceptions that I have
proposed are of exactly the same nature. I claim, for example, that in
order to _control_ a configuration of a constant kind, it is always
necessary to _vary_ one or more sensations, and that to _control_ a
sensation of a constant kind, it is necessary to _vary_ one or more
intensities. In each case, the presence of disturbances is assumed.
Likewise for all the other levels I propose: for example, in order to
_control_ a principle (to act in such a way as to keep it true in
perception), it is always necessary to _vary_ some programmatic processes
(reasoning, prediction, logic, rule-following, etc.).
As always, the "evidence" that is needed is a report that any honest
observer can see the required relationships among variables. It doesn't
matter in the least whether the observed variables are inside or outside
anyone's skin; there is no requirement for _simultaneous_ observation by
different observers; only that different observers acting independently
will report the same phenomenon. If simultaneity were a requirement,
scientific journals would be useless. Science is about achieving agreement
among observers, after each observer has independently examined the world
of experience and acted upon it as prescribed. If I report that sticking a
pin into my hand hurts, and you report that sticking a pin into your hand
hurts, and all honest observers report the same thing, we can take "hurt"
as an established phenomenon, and sticking a pin in a hand as an operation
that will produce the phenomenon -- all this despite the fact that only one
person can observe each hurt.
Now, what about explanations? Given that we have established the reality of
a phenomenon, how do we go about explaining it -- stating why it occurs as
observed? We now leave the realms of observational evidence and enter the
world of modeling.
To say why a phenomenon occurs in the sense of explaining it does not mean
asking for its higher-level motivation (why did I drive the car downtown?
Because I needed to buy a computer). The why we are after in modeling is of
the nature of the answer to the question, "why does the light turn on when
I flip the switch?" We are asking about underlying mechanisms which, if
they really existed, would make the phenomena we observe into logical
necessities. If the switch is wired to the light-bulb, and all the laws of
electricity are true, and all other necessary conditions have been
established (for example, there has been no power failure), is it
_necessary_ that the light become bright when the switch contact is closed:
nothing else could possibly happen under the given conditions.
That's how a model works. If the model corresponds sufficiently well to the
underlying reality, then the phenomenon we experience under given
conditions MUST occur. It is a logical necessity, meaning that for it not
to occur would amount to a logical self-contradiction.
Note that this is not a claim that any model is correct. It is a
description of the nature of a correct model. In an incorrect model, the
model itself is just like a correct model, in that it sets up relationships
making a certain outcome _necessary_. But of course we can set up the
required conditions and find that what the model necessarily does doesn't
happen in reality, or happens differently from what the model leads us to
predict. That's how we tell that the model is wrong: what it predicts
doesn't happen, or happens differently. But it definitely predicts
_something_.
Right or wrong, we can tell we have a true model if the model implies
setting up specific conditions, and predicts a behavior which _necessarily_
follows from the structure of the model and the conditions under which it
behaves.
The PCT and HPCT models are clearly true models, in that they propose
underlying mechanisms from which observable behavior _necessarily_ follows.
We generally present specific models with free parameters; these parameters
allow us to adjust certain details of the model to fit individual
characteristics of the real systems being modeled. But these parameters do
not alter the structure of the model when they are varied. Thus when we can
make a model with a fixed structure fit all individuals' behavior, we have
shown that at least this structure constitutes a valid explanation of how
the tested group of individuals is internally organized.
Finally, what about neural circuit-tracing in the brain? We can clearly
identify control phenomena, and we can clearly propose and test models to
explain these phenomena, without circuit-tracing. If a model works --
predicts correctly -- we know that however the underlying circuitry is
connected, it must accomplish the same end-result that the model
accomplishes, and by means that are at least functionally equivalent to the
means the model proposes.
Indeed, it is usually the case that the model tells us what we are looking
at when we get to the level of circuit-tracing. Anyone who has had
experience with troubleshooting electronic systems knows that without a
schematic diagram, and without an understanding of the model behind the
schematic diagram, it is essentially futile to try to figure out what a
complex circuit is supposed to do -- how it is supposed to behave when it's
not broken. The idea that we could figure out how the brain works strictly
through circuit tracing is sufficient evidence that its proposer has never
done any actual circuit tracing in even a modestly complex system. Even
repairmen for such simple systems as television sets have to go to school
to learn the theory of television before they can hope to diagnose
problems, and diagnosing problems means knowing what you're _supposed_ to
observe in a properly functioning circuit. It means knowing what various
subsections of the circuit are supposed to accomplish -- what they _must_
accomplish if they're working correctly.
Without a reasonably correct model, circuit-tracing is futile. You simply
won't understand what you're looking at, and you'll just be making wild
guesses if you try to go ahead. That's about the state of neurology today.
You must have a model that correctly explains behavior before you can get
anywhere with tracing circuits in the brain. So data about neural
connections is NOT, bu itself, evidence about control.
Best,
Bill P.