[From Bill Powers (950831.0730 MDT)]
Rick Marken (950830.0830 MDT)--
Writing to Hans:
I think it is this attitude that nearly killed the patient who was
attached to your model-based blood pressure control system.
Where did you get the idea that Hans' system nearly killed a patient?
That is strictly out of your imagination. In fact, it was explained on
the net (and in the paper Hans cited) that the system was tested on
pigs, and only after it had been debugged and proven completely safe was
it ever used (safely) with human patients. Did you get a copy of Hans'
paper on this control system? Hans honestly described the difficulties
he had getting this system to work; it's simply not fair to exaggerate
these difficulties and ignore the solutions that were found, and at the
same time invent fairy-tales to make it seem even worse.
Bill Powers showed that your model-based controller controls no
better (and in most cases far worse) than a simple perceptual
control model of the same situation; so your model is neither a
parsimonous nor an accurate representation of the real controlling
done by living systems.
All but one situation: where the feedback signal is lost or part of the
incoming information is missing. Then you need some kind of world-model
to construct the _right_ missing information (or at least enough missing
information to make the system _seem_ to work). We have called this,
loosely, the imagination connection, but without working out all the
details that would be necessary to make such a connection work. Hans'
model (or at least the model he has adopted, I don't know its origin)
tackles this problem and comes up with one potential solution.
My only quarrel with Hans arises from his trying to extend a particular
model that works in a particular situation so that it covers all common
control problems, which it does not. The attempt to deal with
disturbances by modeling them (which is made NECESSARY by his approach)
runs into many problems and the solutions they call for become
unbelievable after a point. Simple and highly predictable disturbances
might be handled, approximately, in the way he suggests. But as we
imagine more realistic kinds of disturbances, the machinery necessary to
predict the immediate course of these disturbances becomes more and more
complex, and we very soon, in my opinion, pass the bounds of parsimony
Your own comments support the idea of model-based control at the higher
levels of organization. You form a picture of some person you have never
met called Hans Blom. This model begins with nothing but some statements
in the English language which describe, roughly, a specific approach to
model-based control (which you may recall had been discussed in a
preliminary way by us years before we heard of Hans Blom or the model he
describes). Since these statements are incomplete, you fill in what you
imagine to be other properties of Hans Blom, such as his thoughts,
beliefs, motivations, attitudes, biases, understandings, and so forth,
until you have constructed a fairly complete model of a human being.
This model makes sense of other things Hans Blom says, and can be
infinitely modified until it seems to explain everything about him. You
then behave to control your relationships to this model of Hans Blom,
just as if the model were the real person.
This is how we invent heros and villains. We take a few rods and joints
to erect a child's drawing of a stick man, and then invest this armature
with imaginary properties that satisify our wish to worship someone or
hate someone. The materials with which we finish the construction come
from our experiences with people we have admired or hated, bits and
pieces of interactions with them, and wishes and fears about what might
be true of them and ourselves. When we look at the actual people on
which this model is built, we do not see what is before our eyes; we see
our mental models.
Tom Bourbon [950830.1311] --
You've put your finger on a major problem in modeling: mistaking curve-
fitting for a generative model.
What these physicists did was to impose a rule such that the particles
moved in the average direction of the particles around them, plus a
random variation. Behavior which fits this rule also fits the
implications of the rule. However, it might also fit the implications of
many other rules that could have been imposed.
It might be useful to notice that particles seem to follow a rule like
this, just as it is useful to notice that E. coli seems to follow a rule
that says "generate a new random direction of swimming if the direction
of swimming is down the gradient." The implication of that rule for E.
coli is, interestingly enough, that it will progress up the gradient.
However, what we really want to know is how the particles, or E. coli,
might be organized inside so they would seem to follow the apparent
rule. Obviously, E. coli does not contain a little thinking circuit that
somehow knows the direction of its own swimming and the direction of the
gradient outside it and that follows this logical rule. Whatever is
inside E. coli gives the _appearance_ of following this rule: that is,
if we did follow this rule in manipulating particles, the result would
be the same as the behavior of E. coli. But this by no means shows that
the behavior of E. coli is driven by that rule, or any rule.
In fact, Koshland found that E. coli simply controls a signal that
derives from the speed of swimming and the cosine of its direction
relative to the gradient. E. coli has no way of sensing the gradient
itself or its own direction of swimming in laboratory space and it
certainly can't compute cosines. It would seem extremely unlikely that
E. coli would have any conception of three-dimensional space, position,
or velocity vectors. So it couldn't possibly be obeying a rule in which
the inputs are given in such terms.
The actual explanation is much simpler. In logical terms, the "rule" is
"if negative signal, tumble." That results in behavior that exactly fits
the behavior predicted by the more complex rule given in terms of actual
gradients and spatial directions. Actually the relationship inside E.
coli is not logical, but analog: the less positive the input signal, the
shorter the interval between tumbles. Koshland verified this by
adjustments of the rate at which perfused concentrations changed. So the
actual mechanism doesn't employ logical rules at all.
You can always find an apparent rule that governs any behavior. Have I
reached the ground yet? If no, continue falling. If yes, SPLAT!. This is
the rule that accounts for the behavior of raindrops. If you move model
raindrop particles according to this rule, sure enough they will fall to
the ground and splash. Moreover, model raindrops will show a strong
tendency to move in parallel directions, according to this rule, and
when you look at the real raindrops, they are indeed falling parallel to
each other! What more proof would you need that this is the right model?
Avery Andrews (950831.1930) --
If most disruptive schoolchildren didn't actually have higher-level
goals that could be satisfied by getting along and doing well in
school, and satisfied better that way than by disrupting, they
couldn't be reached by a fundamentally non-coercive program; the
only workable program would be a gulag to confine them in.
Look at that from a different angle. Non-coercive schools are a
relatively recent concept and people have considerable difficulty in
giving up their old approaches to try the new one. On the other hand,
coercive methods for solving just about all human problems have been
under experimental investigation for something like 10,000 years. While
we can't yet say that a non-coercive approach will solve all our
problems as a human race, we can certainly, by this time, say that the
coercive approach has produced a mess that is rapidly getting worse.
Of course this is a contingent assumption, it might be false, and
there's lots of apparent counterevidence around, especially in
places like Rwanda and Bosnia. On the other hand, if it's false,
the future outlook for humanity is grim, given the ever-increasing
loop-gains of human control-systems.
What we see in Rwanda and Bosnia is the natural outcome of the coercive
approach to human relationships. It is the idea of continuing that
approach that makes the outlook for humanity grim.
The irony of the situation is that the people who use reward and
punishment and want everyone to conform rigidly to the local rules think
that doing so will make things better for themselves. That is why they
do it, to make things better. When the children they raise or teach turn
out to believe in the same things, the people doing the raising and
teaching pat themselves on the back because at least a part of the next
generation will have the same values and use the same proven methods. It
never occurs to them that things are getting worse BECAUSE they are
raising children this way -- that they are simply assuring that the next
generation will have the same problems, only worse. The next generation
will grow up saying "Well, it worked for my parents; if we could all
just go back to doing things the way they did, we wouldn't have the
problems we're having now. Just look at the violence we have today; my
parents never had to live through that." What they don't realize is that
today's violence is largely a result of what yesterday's parents made
sure to teach their children in the hope that the bad conditions of the
parents' day could be restored to the better conditions of the
grandparents' day. Nobody lives long enough to see that continuing and
improving the same social practices is gradually escalating the
I don't believe in forecasting the future; I believe in changing
intentions and thus changing the future.
Best to all,