Theory vs theory, or theory vs observation?

[From Bill Powers (940503.1000 MDT)]

Hans Blom (940503) --

OK, you asked for it.

I did indeed, and thanks for delivering. Since I haven't been told
to get off the net yet, I will feel free to respond.

2. Therefore, any model is a subjective thing, which depends
upon one's taste about what one considers important and what
not. As control engineers are apt to say: one man's noise is
another man's signal.

While at a meta level I agree with you, you're leaving out the
subject of phenomena. Models, as I think of them, are meant to
explain how a system can produce the phenomena we observe. The
phenomena I've been interested in have been the really simple
aspects of human and animal behavior, the same things the
psychologists have been working on for a century or so. Without
phenomena to constrain them, models can take on any form, and as you
say, which model or models you adopt is then pretty much a matter of
personal taste.

But the kind of models I respect are those that explain and predict.
This means that not all models are equal. Some predict better than
others, and some require adopting premises that are more or less
verifiable than the premises of others. So it's not totally a matter
of taste (except, of course, for the personal preference for models
that behave like the real system and involve realistic premises).

3. Therefore, proposing a model to others is very much the same
thing as saying: "Hey, look, this is what _I_ think is
important". Proposing a model is a social activity, an attempt
to focus another person's attention on the things that you see,
one of the steps in the creation of a common language.

True, but I try to focus attention on aspects of behavior that are
simple enough that we can easily agree on what needs to be
explained. I didn't start with a model and then try to find
behaviors that might fit it. I started with behaviors of common
kinds, about which there is little disagreement, and then tried to
construct a model that would handle as many of the details as
possible. One of my criticisms of other models is that many of them
don't handle the details and don't predict properly. This isn't just
a matter of saying "My model is better than your model." It's a
matter of comparing models with the phenomena they're supposed to be
about, and seeing that some ideas proposed as models don't actually
explain what they purport to explain.

4. As a corollary, rejecting another person's (correct) model
is very much the same thing as saying: "What _you_ think is not
important". Or: "Speak my language. I am not willing to speak
yours".

"Correct" can be taken two ways. Lots of models are correct in terms
of their internal logic; that is, they don't contain any
mathematical errors. But models also have to be judged in terms of
how well they fit and predict observations, and there even the most
impeccable internal logic can't substitute for correct predictions.

5. Proposing one model and rejecting another (correct one) is
therefore a power game.

Yes, I agree. So we should focus on what we mean by "correct," and
not let arguments rest only on personal whims. The best situation is
where both models are about the same phenomenon, both make
predictions about observable relationships, and only one model
predicts correctly. Then it's easy to decide between the models.
Doing it this way removes the "power game" aspect, unless one party
considers appealing to observation to be a form of power play. I
have spent a lot of time thinking up experiments and demonstrations
that will eliminate certain classes of models, not because I dislike
them but because they don't work.

It is this attitude of "I'm right, you're not" or "I know
what's important in life, you don't" that so often wreaks havoc
in csg-l circles.

It's hard to keep personal frictions separated from scientific
controversy: to avoid confusing "this model is right" with "I'm
right" and "that model is wrong" with "you're wrong." Unfortunately,
all of us to some extent attach our personal feelings of worth to
the models we espouse, so that an attack on a model is felt as an
attack on one's person. But isn't it the business of science to try
to decide among different models of the same phenomena? Sooner or
later, the less capable models are going to fall by the wayside, and
it's pretty much up to the individual scientist to sort out the
personal feelings involved in having chosen the wrong model. Poor
old Priestley went to his grave insisting that phlogiston theory was
really the right model for combustion, and I suppose that others
regretted his discomfort at having lost the battle, but nobody
regretted giving up that model.

PCT and its methodology certainly conflict with what many
older theories have assumed to be true.

As I said above, this is generally true of different
models/theories. It is to be expected. Why else a new theory if
it does not want to focus the attention on something not
included in other theories?

But PCT conflicts with certain older models about the explanation of
_specific phenomena_. If different models simply cover different
phenomena, then they can coexist independently, but when they're
about the same phenomenon, and are required to predict that
phenomenon, then something has to give way. The way you're talking,
we should never throw a theory away -- just keep accumulating them
indefinitely. This isn't how science has ever worked before.

... when a external agency comes along and
starts pushing you toward one of the conflicted goals, all
doubt seems to disappear: you take the side of the other goal
and wholeheartedly push back.

So? You see the same thing in a well-balanced heater/cooler
system with a common setpoint where are both operative.

True, but I'm not talking about a well-balanced system. Conflict
occurs (as I define it) when two control systems simultaneously seek
incompatible goals, like a heating system trying to maintain a set
point of 72 degrees and a cooling system trying to maintain (in the
same physical location) a set point of 62 degrees. One example of a
conflict is between explaining behavior as being caused by the
environment, and at the same time trying to explain it as being
caused by internal intentions. A person who has accepted the former
theory, and has begun to assimilate the control-system model, might
well realize that these explanations are incompatible in specific
situations, yet find some sense in both of them. My experimental
demonstrations have been designed to sharpen conflicts like this,
reducing them to a situation in which both explanations obviously
can't be right.

     A small excursion into the philosophy of science. One of
     the correlaries of G"odel's work is the discovery that no
     single theory can cover all of truth. The best thing a
     theory can do is to build a "web" to "cover" the universe
     of truth. But inevitably every theory must leave holes. In
     turn, these holes may be covered by a web, but again holes,
     now smaller, will be left. In essence, G"odel's work
     states that even an infinite number of theories will not be
     enough to fully comprehend everything. Given our human
     limitations, it is not so bad to be eclectic and collect
     "pretty" theories.

I think you overstate the scope of Godel's theorem, but I don't
disagree that no theory has ever, or is ever likely to, explain
everything. I have never claimed that PCT explains everything, and
in fact have often gone out of my way to list things it doesn't
explain. What I am concerned with is how well PCT and other theories
explain _specific_ aspects of behavior, where they both are talking
about the same phenomenon. If a theory fails to explain phenomena
that are specifically within its stated scope, there's something
wrong with it, quite independently of what any other theory might
say.

An "assimilation" is impossible because models will generally
be in conflict. I know of no model that builds upon another
while leaving the older theory intact.

Einstein's theory left Newton's theories intact, within the range of
small velocities, as NASA will attest. PCT leaves parts of S-R
theory intact, if we understand that "stimuli" are really
disturbances of controlled variables. Models that are truly in
conflict can't coexist: they make opposite predictions about the
same phenomena. Some aspects of S-R theory are in conflict with PCT,
and of the two, only PCT makes correct predictions in those regards.

Almost correct, but not quite. Some have said: "Hey, great, I
can use that!". Others have said: "You presume too much. Even
if you have something worthwhile to say, I will not tolerate
your attitude of pronouncing everything that I previously
learned as insignificant or erroneous. Who do you think you are
to invalidate me and my whole life thus far?"

When you put it that way (and evidently there are significant
numbers of people who do) the answer is that I have no right simply
to pronounce another person's work and life wasted. But sometimes it
is a fact that a person can waste many years on a wrong idea. It is
a very common occurrance. For people who can admit that this has
been true, the experience is unpleasant in the extreme, and one can
only hope that what comes next will make up for the trauma.

However, the protest "But if that's true, everything I have learned
and believed is wrong" is no argument against a new idea. It's only
an explanation of why a person might resist the new idea. It is also
no argument against proof that the old idea is wrong even in the
absence of something to replace it. Every scientist quite naturally
dreads the threat of having some basic premise of his argument
destroyed by the facts. But a real scientist does not use that as an
excuse for denying the facts or clinging to an old idea long after
it has ceased to be defensible.

                    Now the interesting thing to me is that
these arguments have had no common theme; that is, they
haven't revealed any consensus to the effect that there is
some particular basic flaw in PCT.

Is the reason for this clear now?

No, it isn't. What you've offered as reasons have to do with
theories that cover different ground, and therefore can't conflict
with PCT, and simple psychological reactions to the threat of having
been wrong. Where I have noted disagreements between PCT and other
theories, these disagreements are direct conflicts, requiring
certain propositions to be both true and false at the same time. My
point was that in none of the arguments about these points of
conflict has the conflict been resolved by showing a mistake in PCT
-- a wrong prediction based on a correct application of PCT as it is
stated. In every case I know about, the mistake has been in the
other theory, in its prediction of something that does not happen.
This is not a matter of personality clashes or mere preferences. It
is a matter of which model reproduces natural phenomena the most
correctly.

The problem is the exclusivity that PCT claims as the one and
all theory, the theory to end all other theories as
insignificant. This is not acceptable to a scientist.

I repeat, I have never made such a claim. I have always limited any
claims about superiority to those situations in which we have tested
the model against real data using real people, and have found that
the model predicts correctly while rival models expected to deal
with the same phenomenon predict incorrectly. Most of my criticisms
of other models are not made in comparison with PCT: they are
criticisms based on what the other models say in comparison with
real phenomena. I criticize, for example, the use of population
statistics as a way of deriving individual characteristics. The
flaws in that approach can be laid out without any reference to any
other approach. The same is true of any theory which claims that
precalculating actions is a sufficient basis for producing
consistent effects of actions in the natural world. Those claims
have been known to be wrong for over a hundred years -- yet they are
still made. And they are wrong on their own merits, not just in
comparison with PCT.

In the hardest science of all, mathematics, for example, we
have both Euclidian and non-Euclidian geometries.
Mathematicians do not fight about which is the "correct"
geometry. They know that both, although in full mutual
conflict, are correct _in themselves_ and that both are
extremely useful tools, al- though for different types of
applications.

There is no conflict between Euclidian and non-Euclidian geometries.
Each holds true under different conditions: planar versus non-planar
space. Non-Euclidian geometry reduces to Euclidian geometry in the
case of zero curvature; Euclidian geometry holds only for
relationships in a plane. A true conflict is a contradiction, a
requirement that proposition A be both true and false at the same
time, under the same premises. A true conflict between theories is
seen when under the same conditions, the theories predict outcomes
that are physically incompatible: for example, a variable must
increase according to one theory, and decrease according to the
other. It can't do both.

See? What others deem important is not so to you. Mutual
inconsistency of theories is not the problem, as I hope is
clear by now. The basic problem is the presumptuousness of PCT
as the one and only theory that saves the world. Very, very
sectarian.

Mere mutual inconsistency of theories is never the problem, I quite
agree. But theories must be consistent with observation,
individually. Where two theories predict incompatible observations,
one of them (or possibly both) must be wrong. They can't both be
right: that is, when they make incompatible predictions, the only
choices are that neither prediction is upheld, or only one of them
is. The quantum theory of electron propagation is incorrect under
the two-slit experiment, while the wave theory is correct. Either
you get interference fringes, or you don't: you can't both get them
and not get them.

Under S-R theory, a disturbance added to output effects downstream
from the output will simply cause the outcome to vary exactly as
predicted from the disturbance. Under PCT, the same disturbance
added in the same way will cause the output, upstream from the
disturbance, to vary, while variations in the outcome remain many
standard deviations less than those predicted from knowledge of the
disturbance. That is a direct contradiction, and only one theory, at
most, can be right.

I don't buy the eclectic approach; it leaves out of consideration
testing theories against nature. It's perfectly possible that of all
the theories offered to explain a given phenomenon (like
consciousness, for example) there isn't one of them that's worth
paying attention to.

Isn't it generally accepted that if a theory doesn't give any better
predictions than you get by flipping a coin, the theory should be
discarded?

Because in its presentation PCT is a dogmatic power game. "Bill
Powers is right and Rick Marken is his prophet".

Now who's playing the personality game?

You do not play by the rules of science. As long as you believe
that another theory is wrong simply because it is in conflict
with PCT, you are so unscientific as to be unwelcome in the
circles where scientist convene.

I think every scientist deserves one consideration before all
others: that in criticizing a scientist's theory, one should learn
what it is. I have spent many hours going over Rick Marken's posts
to see what people have objected to. While he has produced some
outbursts that would better have been left unsaid, by far the
greatest part of his posts has been concerned with making clear the
structure and logic of PCT, the way in which predictions are made
from PCT, and the way PCT is tested against observations. Most
arguments with others have arisen from their attributing to PCT
statements, logic, and conclusions that are not part of PCT. So most
of what Rick has been railing against consists of straw-man
arguments, arguments that show only that PCT has not been
understood. As with most theories, the only way to really understand
them is to learn how to play the game according to the theory.

If you can find one post from RIck, me, Tom Bourbon, or any other
old hand at PCT that puts down another theory "simply because it is
in conflict with PCT," I'll grovel in abject apology. The only
conflict that any of us is concerned with is a conflict between a
theory and the facts it is supposed to explain. We have seen PCT to
be in close agreement with such observations as we have been able to
make. And we have seen that other theories which we reject are NOT
in agreement with observation. That, and not a simple comparison
directly between PCT and other theories, has been our sole basis for
claiming that PCT works better than those other theories. That
doesn't seem unscientific to me; on the contrary, that is the
essence of the scientific approach to theory, as I understand it.

Looking over this post, I guess I would summarize it by saying that
you seem most concerned with a theory-to-theory comparison, and that
I am most concerned about the middle term missing in your arguments:
the observations that the theories are supposed to explain.

ยทยทยท

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

Bill P.