[From Bill Powers (950904.0810 MDT)]
Marc Abrams (950903.0830) --
I've been sort of coasting along and shooting my mouth off, trying to
get a feel for where you're coming from. I guess when I sense a conflict
with someone over something, I probe around more or less automatically
to try to gauge the shape of whatever I'm up against. It's the Test for
the Controlled Variable, although I can't claim that it's a conscious
thought-out strategy (it's easier to see afterward than during). You
seem to do something similar. It's a fairly irritating strategy when
carried too far, judging from the results.
One of our disagreements seems to be over whether there are any actual
facts about human behavior, independent of who is looking at them. I
think we would agree that there are no facts in some universal objective
sense, at least not that we know about. But there is another, lesser,
kind of fact, a fact based on observations and reasoning that anyone can
check out and either verify or find false. One example is the inverse-
square law of gravitational attraction that Newton proposed over 300
years ago. It happens that later work has shown that this isn't the
whole story of gravitation, but it's close enough that anyone who wants
to check it out under ordinary laboratory conditions will still be
forced to agree that Newton had it right as close as we can measure.
It's interesting that this kind of checking does happen and does work;
it tells us that even if we don't know the absolute Truth, what we're
doing has some connection to it.
The facts of control theory are that kind of facts. We can do simple
experiments, some so simple that they involve only two people and two
rubber bands, that illustrate the principles of control, and anyone who
does these experiments will see the same relationships and the same
principles at work. Of course the PCT interpretation is not some magical
window to the Truth, but it is so tightly connected to processes that
anyone can observe that it becomes very difficult to think of any other
equally simple interpretation that works anywhere near as well. So in
the same terms that we judge Newton's laws, we are justified in saying
that PCT is based on facts about control behavior.
There's another layer of facts that is build on the first layer. These
are the facts that logically have to be true if the first layer is true.
For example, we may observe (at the first level of facts) that in a
control process something is being maintained in a constant state, or in
some repeating pattern of change. We see that this happens even when
disturbances are being applied to whatever is being controlled. We see
that the actions of the controlling system automatically differ from the
undisturbed case in just the right way to cancel the effects of the
disturbance and keep the controlled variable steady or changing in
nearly the same pattern as before.
So now we can ask a question: what would happen if two control systems
like the one we're observing were controlling two variables in the
environment, and these variables were very tightly coupled together?
That is, what would happen if changing the state of either variable
would necessarily have a strong effect on the state of the other
variable?
From the standpoint of either control system, the actions of the other
one would just be a disturbance, tending to change the variable being
controlled. The first system, if it continued to work as before, would
vary its actions in any way needed to keep its own controlled variable
steady or changing in the same pattern. And of course, the second system
would do the same thing, if it too continued to work as before.
This we know just from the first-level facts we have already observed
about control behavior of single systems. Now, however, when we imagine
two control systems together under these conditions, we can deduce what
has to happen strictly by reasoning it out, without any more
observations of real systems. If each system reacts to the actions of
the other system, it will increase its efforts as much as necessary to
keep its own controlled variable in the original state, the state the
system wants. But BOTH systems will do this. The harder the first system
tries to counteract the disturbances from the second system's actions,
the harder the second system will try to counteract the disturbances
from the first system's actions. If the systems are equally sensitive to
errors and have an equal ability to produce output based on errors, the
result will be that the controlled variables will be frozen in place,
while the actions of the two systems are raised to the maximum possible
level, cancelling each other out and accomplishing nothing. It's obvious
that violence is a natural consequence of conflict.
Those are the second-level facts about conflict. As soon as we come up
with a second-level fact like this, deduced from first-level facts, we
want to see if we can make the second-level fact into a first-level
fact. We look around for situations that seem to fit the conditions we
imagined, and see whether they agree with our deductions.
We might, for example, observe two arm-wrestlers of equal strength. The
hands of the two arm-wrestlers are tightly coupled together; neither
hand can move without moving the other. Each arm-wrestler wants the
other's hand to be in a different place, one against the table to the
north, the other against the table to the south. The actual result is
that the locked-together hands go to neither goal position; they wobble
back and forth between the goal positions, while the efforts of each
wrestler are the maximum that each wrestler can produce. Both arm-
wrestlers have lost control of the positions of their own hands, because
their control systems' outputs are cancelling each other.
When we've made enough observations like these, and done enough
experiments where we can measure what happens, we can say that these
predictions about conflict are now first-level facts. We can describe
experiments that anyone can do, and anyone who does them will come up
with the same result and will agree with the same explanation.
By checking out the second-level facts, we do more than just prove that
we were right in our deductions. What we usually find, as we would in
this example, is that our predictions are right only under some
conditions. We find many conditions that result in failure of the
prediction, and from those we are led to go back to the first-level
facts to fill in things we hadn't considered, for example the capacity
for continued effort by the muscles right after a previous arm-wrestling
match.
We can also be led to realize that there are third and fourth level
facts, and more, that can make a difference. If we get an NFL linebacker
to arm-wrestle with his wife, we might observe that the wife always
wins, even though the line-backer can win against anyone else on the
team. If we were naive, we would conclude that line-backer's wife could
therefore also win against anyone else on the team, but an experiment
would quickly put the kibosh on _that_ prediction. What we have found is
that the linebacker doesn't _want_ to win over his wife. He knows he
could win but he doesn't want to humiliate her because he loves her. All
that mushy higher-level stuff is very difficult to verify under the best
conditions, but it can have a radical effect on the results.
It's easier, however, to go the other way. If we observe two people arm-
wrestling and see that neither is winning, we are usually justified in
seeing this as a conflict situation. We have ways to test to see how
hard each party is trying. We can also model each party individually in
an experiment where artificial disturbances are applied, and come up
with a computer simulation that will match what the real person does. We
can do this for each arm-wrestler without the other one present, and get
two sets of first-level facts. Then we can run the two models together,
coupling the outputs together as the real ones would be coupled, and see
what the model predicts. And finally, we can couple the two arm-
wrestlers together and see if they behave as the model does. If all this
checks out, we can be pretty confident that we are seeing a true example
of conflict. A good part of Tom Bourbon's experiments with multiple-
person control tasks works just like this, although not with arm-
wrestlers.
So while we can't be sure that the second-level fact will always be
true, we can lay out a public procedure that anyone can use to see if
it's true in any given case. "True" means, of course, "In agreement with
the observations under an interpretation that's the best one we can
think of." And the emphasis is on "we": what one person finds to be the
best interpretation isn't enough: it has to be an interpretation on
which independent observers can agree. In the best possible outcome, all
independent observers will think up independent interpretations which
turn out to be the same.
OK, this is getting pretty long but I want you to see where I'm coming
from. The definition of conflict in PCT is very precise; it's based on
experiments that have been done and working models that have been
matched to two-person and up to four-person experiments. As near as
anyone human can know, we know some facts about conflict.
However, we also know that we are dealing with simple situations and
that more complex ones exist. For the present discussion, the most
important complexity that remains to be investigated is the case where
conflict is multidimensional. Few real situations involve the control of
just one variable in one dimension. People control many variables at the
same time; when their behavior interferes with what other people are
controlling, the interferences can exist in many dimensions at once.
We are still at the stage of proposing second-level facts about these
complex situations, and haven't yet started the work to try to convert
them into first-level facts. But there are some things we have
discovered by logical reasoning and by testing ideas in computer
simulations. One conclusion on which I think the modelers will agree is
that it's quite a lot harder to produce true conflict in a
multidimensional situation. The exposition to follow, if you will be
patient, may provide a basis in PCT for the difference you see between
competition and conflict (although the resolution may not be quite what
you expect).
We might as well go directly to a market example, since that's what
we've been arguing about. Let's consider two shops in Durango that sell
gift items to the tourists. From the simple-minded arguments I have been
using, it's obvious that there is a conflict between these shops; each
is trying to sell things to the tourist population, and the money one
shop gets out of a tourist is no longer available to be spent in the
other shop.
However, let's reconsider what it is that the tourists find when they go
into the shops. As you say, it's futile to try to be all things to all
people; no matter what selection of items is in the shop, there are
tourists who will leave without buying anything, even if there is no
other shop. Neither shop can carry all the items that any random drop-in
customer might want.
If both shops carried exactly the same items, there would have to be a
conflict. We can forget the tourists who don't want any of the items;
they wouldn't buy in either store. But if there are tourists who want
wood-carvings of Swiss yodelers that contain blue glass ashtrays, then
if they bought them in one shop they would NOT be buying them in the
other shop. There would be direct competition to sell the same items to
the same people.
So what happens instead? One shop sells Swiss yodelers and paintings of
elk on black velvet, and the other sells ceramic dolls with huge eyes
and whoopie cushions. What happens is that the shop owners try to
_avoid_ direct competition, at least with respect to the main items that
might bring people into the shop.
This is what we find in modeling multidimensional control processes,
where _internal_ conflict is possible. In a self-adapting collection of
systems, the multiple control systems that share a common pool of
controllable variables end up using all the same variables for their own
purposes, but with different weightings so that each one can control a
different combination of variables while minimizing conflict with all
the other systems. At a higher level, each one controls the aspect of
the environment that it perceives while all the others do the same, but
the "aspects" are build on different weightings of the underlying shared
variables. It is actually possible, under conditions we can describe
fairly accurately, for the control systems at the higher levels to act
without any conflict with the other systems, or with so little conflict
that no system is rendered ineffective.
The case of the Chinese restaurants illustrates the point. There were
four in Durango, each offering rather similar food, but with significant
variations: one was fast and cheap, another was slow but had a bigger
menu, another offered Korean food, too, and the fourth was more
expensive but had better food. Then a fifth restaurant opened much
farther from the center of town. It carried all the same food that the
others carried, more or less; it was cheap and fast; it offered some
non-Chinese food; and it had some gourmet specials at higher prices. It
went out of business because it was competing directly with four other
restaurants, and just being equal to them in their various specialties
wasn't enough to steal enough customers away. At least that's my guess.
In technical terms, a given environment has only a certain number of
degrees of freedom in which it can be controlled by the actions we can
produce -- or rather, only a certain number that can be independently
controlled at the same time. If the number of independent controllers is
equal to or less than the available degrees of freedom, it is possible
for the controllers to select independent mixes of environmental
variables to control, with exactly the right mixes resulting in total
freedom from conflict. However, if there is just one more control system
than degrees of freedom, conflict is inevitable. The last system to be
added, the surplus one, finds itself in conflict with ALL the other
systems no matter how it tries to adjust the mix of variables it is
controlling. I think that could be what happened to the fifth Chinese
restaurant.
When you said that no company could be all things to all people, and
that companies looked for the needs of customers that they could meet,
you were implying a certain degree of flexibility in what a company
could claim as its product. A company will become successful when it
finds a mix of products that fits a certain segment of the demand.
But the unspoken aspect of this is that finding this particular
successful mix is largely a matter of discovering potential conflicts
and avoiding them. If a company finds itself in direct item by item
competition with another company, then everything I said about
competition would be true: for one company to win, the other must lose.
All successful businesses learn, whether they want to or not, that this
is a situation to be avoided unless what is wanted is an all-out war. An
all-out war will most likely put one company out of business and
possibly both. A company will enter into such a war only if it is sure
it will be the one that is left operating.
You will often find a Texaco gas station close to a Conoco or a Marathon
station. But you won't find two Texaco stations next door to each other.
The different companies offer different kinds of values; final filters,
super duper high-test unleaded with valve cleaner, convenience stores,
car washes, 10% alcohol, Mechanic On Duty, or cheap tires on sale. They
announce these wares prominently in ads and store displays. The parent
oil companies go to great lengths to give each franchise an assortment
of features to advertise and display that are as different as possible
from those of other companies. The point is not to compete, but to avoid
competing with the other companies. The one variable that would lead to
direct competition, price, is never mentioned in ads, and not only
because prices can change. Price wars have killed off enough gas
stations to make the operators avoid them by any means possible,
including collusion. Let just one station operator advertise gas for 10
cents per gallon less than the next guy, and disaster is around the
corner -- or arson.
What the free market system offers is an opportunity for an entrepreneur
to find a niche by offering a mix of products and services that
satisfies the needs of some part of the population. But the
entrepreneur, to be successful, has to avoid direct competition as much
as possible. It isn't competition that makes the system work; it's the
ingenuity of people in finding combinations of goods and services to
sell which, ideally, are not to be found anywhere else -- certainly not
nearby.
At another level, of course, the competition is always there. It can be
avoided only if the number of independent businesses remains below a
certain number. The staggering failure rate of new businesses, however,
shows that there are always more entrepreneurs trying to enter the
market than the market can sustain. There are not enough degrees of
freedom that actually matter to people to let everyone who wants to
start a small business (or probably a large one) find a niche without
coming into direct competition with another business that already
occupies an almost identical niche. And at a still higher level, the
Composite Consumer can spend only as much as it earns (in the long run),
and that is the pie that has to be divided among the entrepreneurs.
Most of what I've been saying here is built on second-level facts and
until someone does the required research can be argued with. But I think
that the principles of control theory do have an application in economic
matters and could even teach us some things that haven't been understood
clearly before. The main new notion here is that the health of the
business community depends on the avoidance of direct competition,
because direct competition is simply conflict, and conflict prevents
control systems from working properly. There are many other
considerations involved in economic health, but I think this is a major
one.
···
-----------------------------------------------------------------------
Best regards,
Bill P.