[From Bill Powers (930316.1500)]
Hans Blom (930316) --
The second of your adaptive control diagrams is essentially the
same scheme I was proposing a while ago for model-based control.
As you say, model-based control would have to use some systematic
algorithm for making the model's response agree with the
environment's -- not a matter for random reorganization.
I don't think that model-based control is very practical for real
behavior unless there is also a provision for making corrections
in the action based on the deviation of the model from the real
environment's response. All that the model-based scheme can do
for you is carry the system through periods when real-time
feedback is unavailable. This, of course, can be critical, but no
organism could rely on it for long. There has to be some way to
reset the model at frequent intervals, because it will always
drift away from reality rather rapidly. Just think about walking
around in a completely dark room after a power failure at night.
You can go a short distance without running into anything, but
you can't go far without at least touching something identifiable
to reorient your imaginary model of you in the room. It's
interesting how little tactile information is needed to
accomplish the resetting, but without any such information at
all, no mental model stays in calibration for more than a few
minutes (if that long). And of course even the best mental model
fails immediately if the environment is disturbed.
Your response to Rick Marken:
This must have to do what you earlier called "the central
insight of PCT". This position is not revolutionary, however.
Bishop Berkeley already exposed the view that sensory input is
a dependent variable in his "A treatise concerning the
principles of human understanding", published in 1710. His
philosophical position is called a strict empiricism: "esse est
percipi", to be is to be observed. His position leads to
propositions like "the tree that I see over there exists
because I see it".
If you think this central insight is philosophical or mystical,
you've missed the point (your quoting the good Bishop makes me
sure you have missed it). PCT says nothing about the existence of
the tree over there. Let the philosophers worry about that. All
that PCT says is that you can't avoid bumping into the tree
unless you can see your relationship to it. That means seeing it
and its distance from you. If you're wearing trick glasses, you
may bump into the tree anyway, because you can control only the
relationship you perceive, which may be different from the one
that actually exists. If the tree is really a holographic
projection, you'll still steer to walk around it because you can
control only what you perceive; as far as your perceptions can
tell your brain it's a real tree and walking into it will hurt.
All that the brain knows about the external world comes to it in
the form of perceptual signals in the afferent neural pathways.
There is no other way for that information to get into the brain.
If the brain wants to control the position of a real glass on a
real table, it's out of luck: it doesn't have any way to know
about the real glass and the real table. It can, however, adjust
its output signals so that a neural signal representing the glass
can be manipulated to achieve a certain relationship with a
neural signal representing the table. THAT the brain CAN do.
I should think that all this would be self-evident to any
engineer who has ever actually built a working control system. A
real hardware control system can't interact directly with the
physical plant it is controlling. All it can do is alter its
electronic output signals and see what happens to the signals
being generated by its sensors. That's all it knows about what is
happening outside it. If the sensors jump out of calibration, the
control system will happily continue controlling the
miscalibrated perception, while the technician in charge rushes
to hit the STOP button. What is controlled is ONLY what is
perceived. One hopes that what is perceived has some relationship
to what IS, but that is something that has to be determined
indirectly.
This simple concept which should cause no problems for any
control engineer causes immense problems for conventional
sciences of life. The reason is that these conventional sciences
ignore the difference between what is perceived and what is -- at
least when they're trying to explain behavior. And not having any
experience with real system design, it seems perfectly reasonable
to such conventional scientists that a stimulus input from real
objects in the environment should be able to cause motor outputs
that steer the organism through a variable environment along a
path to the cheese or the mate or whatever. What's the problem?
You can see them doing it, so it must be easy.
If you're an engineer watching an organism behave, you will have
a hard time making your mental model behave in this simple cause-
effect way. You will notice that the eyes keep moving around,
that the head moves and bobs up and down, that the steps are a
little imprecise and slightly wobbly, that things in the
environment are shifting around. Being a person who is charged
with making systems actually work, you will wonder how the
organism gets away with such imprecision of action -- where are
all the stimuli coming from that cause the corrections of the
little mistakes and overshoots and hesitations? How does the
environment know that it should stimulate the organism just in
the right way to correct for a previous stumble? How does that
little unevenness in the path send just the right stimulus up the
spine to make just the right muscles change their tension to keep
the leg from jamming into the ground or flailing in empty air on
the next step? Any engineer who pays attention in a professional
way to the claims of s-r theorists would soon walk away shaking
his head. No way!
Unfortunately, engineers seem to abandon their normal
professional attitudes when they start trying to explain
behavior. They start listening to the psychologists and
physiologists and neurologists who think that behavior can just
be "generated," open-loop. Perhaps they're just being polite
because they're on another scientist's turf. They say, "Oh, is
that how it works? OK, you must know what you're talking about;
I'll see if I can make that work."
And of course they can make it work. A good engineer can make any
damn fool idea work. They can build an arm that's as solid as the
front end of a Mack truck, equip it with precision bearings and
gears and stepper motors, compute the driving signals using 80-
bit floating point arithmetic, and make the arm move exactly as
wanted. The smart ones must surely realize that this is NOTHING
like the way a human arm works. But the psychologists see what
they've done, and nod wisely. It works just the way they
expected.
PCT is all about the realization that human systems simply can't
work that way. Their outputs are rubbery and imprecise; their
neural computers are good to maybe 1% at best; they don't sense
everything in the environment that might interfere with the
action. Yet they work precisely and well, for four score years
and six. A person with his little 1% analog computers can get out
of bed in the morning, and perform one action after another all
day long, each action starting where the last one left off, and
16 hours later end up exactly at the side of the same bed, with
no cumulative errors at all. Only one kind of system can
accomplish that sort of behavior: a negative feedback control
system.
I respect your point of view, but if you really believe that
the universe up there at night has no existence independent
from your perception, then there is very little that I can
meaningfully discuss with you: there would be just too little
common language/under- standing.
That kind of respect I can do without. It's condescending.
Actually, I do believe that the universe up there at night has an
existence independent of my perception. I am also convinced that
that universe is nothing like my perception of it. The stars are
not tiny pinpoints against a black background of nothingness. The
Milky Way is not a vague sheet of haze. When Mars slows in its
journey among the stars, reverses its path, and then starts
forward again, I know that my perception is not what is actually
happening. Nevertheless, I can find my way somewhere by keeping
my perception of the North Star in a certain relationship to my
perceived direction of travel, and I know which parts of the sky
to center in my visual field so that Mars will appear in the
middle. Even though my entire education has taught me that the
physical universe is not at all the way I perceive it, I can
still act to make my perceptions behave in ways that have
predictable outcomes, and know that behind the scenes I am having
unseen effects in the unperceived universe that are causing my
perceptions to behave as they do. I can turn a knob-thing on a
radio with full confidence that I will shortly hear voices or
music -- even without knowing what actually happened behind the
knob. Yes, there is a real reality. I am convinced of that. But
it is not the reality I experience. If you gave this ten minutes'
thought, you would realize the same thing.
···
------------------------------------------------------------
Best,
Bill P.