control? attractor basins?

[Hans Blom, 961011]

(Bill Powers (961010.1100 MDT))'s reply to (Martin Taylor 961008)

What one control system does is control. What two do, interacting
in the same world, is not, though what each of them does is indeed
control. The interactions among control system has a dynamic that
almost surely involves attractor basins. ...

On reflection, I'll buy that. ...

Now think in terms of and from the perspective of the cells of our
body and repeat the above. "What one cell does is control," says a
cell. "What two do, interacting in the same body, is not, though what
each of them does is indeed control. The interactions among cells has
a dynamic that almost surely involves attractor basins."

"On reflection, I'll buy that," says the other cell, who feels just
as autonomous.

The question is: what is metaphor and what is reality?

Greetings,

Hans

[From Bill Powers (961011.0600 MDT)]

Hans Blom, 961011--

Now think in terms of and from the perspective of the cells of our
body and repeat the above. "What one cell does is control," says a
cell. "What two do, interacting in the same body, is not, though what
each of them does is indeed control. The interactions among cells has
a dynamic that almost surely involves attractor basins."

"On reflection, I'll buy that," says the other cell, who feels just
as autonomous.

The question is: what is metaphor and what is reality?

This is indeed a problem for any model that treats the behaving entity as a
single complex input-output function with a single highest goal. It is also
a problem when externally defined goals or consequences are not
distinguished from internally defined goals -- when, for example,
"survival" or "efficiency" are thought to be a goal of the behaving system.

In a hierarchical control model, however, one subsystem sets goals for
another, thereby establishing relative levels of control. All systems at a
given level are independent of one another except for mutual disturbances
(which, in an optimally designed system, are minimized). A "higher" system
is not simply a larger collection of the same "lower" systems, created by
redrawing boundaries. It is a physically distinct system, existing in a
different place from the lower systems and communicating via physical
signals with the lower systems.

In other words, the hierarchy is not "molecule, organelle, cell, organ,
organism, tribe ..." and so forth, as many people have defined a hierarchy.
That is only a conceptual hierarchy, not a physical one. The hierarchy in
HPCT is proposed to be a physical hierarchy, in which systems at one level
are made of different cells from systems at a different level, with
specialized cells (neurons) carrying perceptual signals upward and reference
(or parameter-modifying, in a more complete model) signals downward between
levels. The actions of the system as a whole are carried out by sending
signals through neural cells to muscle cells, and the muscle cells contract,
applying forces to the cells making up bones and tissues, through the cells
which comprise the tendons.

The lowest level of behavioral control seems to consist of sensory cells,
spinal motor neuron cells, muscle cells, and tendon cells, arranged to pass
unidirectional effects around in a circle. All systems at this level, and
there are many hundreds of them, exist independently of each other except
for physical interactions due to skeletal and energetic constraints. At this
level the control systems are autonomous, each acting to control its own
sensory signal independently of what the other systems at the same level are
controlling.

However, each system receives a composite reference signal from locations
higher in the nervous system, up to a meter or more away. The higher systems
control variables synthesized from copies of the perceptual signals in the
lower systems (those signal pathways uniformly bifurcate, one branch going
to the local system and the other passing upward to higher systems). The
higher systems consist of neural cells comprising an input function, other
cells that perform the comparison with a net reference signal from still
higher systems, and still other cells that transform the error signal and
fan copies of the result out to contribute to the net reference signals in
many lower systems, including the same systems from which some (or all) of
the original perceptual signals came. These higher systems exist in a
different place from the lower systems; they give the lower systems their
goals, and operate by varying those goals.

It is possible to isolate all systems of the second level, which set
reference signals for systems of the first level and receive input signals
originating in sensory neurons at the first level (some but not all of which
are involved in first-level control). These second-level systems also
receive reference signals from higher systems, and as in the first-level
systems their perceptual signal pathways bifurcate so that a copy of the new
perceptual signals goes upward to the higher systems. For the first few
levels in the brain, this physical arrangement is well-known and
substantiated by neural circuit tracing (although the naming and
interpretations given to the results have varied). Any neuroanatomy text
will show drawings of the pathways. This "reciprocal innervation" is found
at every level of the brain, including the cortex.

Every control system at a given level receives two inputs, one from
perceptual signals of lower levels (or sensory neurons) and one directly
from higher systems that does not involve sensory inputs. The input from the
higher systems is the reference signal. Note that NO reference signal arises
from a sensory neuron. ALL reference signals are the outputs of higher systems.

This is the basic structure proposed by HPCT. It is intended to be not
metaphorical, but physical and neuroanatomical. The evidence is not all
available, to say the least, but so far this physical structure fits what we
know of the geometric, if not the functional, organization of the cells in
the brain (the latter because little is known of functional organization).

And now, the point of this long review of the basic concepts of HPCT.

When we try to extend this hierarchy outside the whole organism, the pattern
immediately breaks down. There is, outside the skin of any organism, no
known higher system which sends reference signals directly into the highest
control systems in the brain. All information from the outside world that
enters the brain must come into it through the lowest level of sensory
inputs, the sensory receptors of the first level. When we consider a group
of people acting together, all we have is the individuals, interacting, like
control systems that all exist at the same level of organization. There is
no physically distinct higher system which receives copies of their
highest-level perceptual signals, compares the result with some higher
reference signal, and sends copies of the ensuing error signal directly to
the reference inputs of their highest-level control systems. The pattern
that holds within each individual simply does not extend outside them.

We can extend the hierarchy downward quite easily. The brain sends reference
signals to every organ, and to the hormone systems via the neurohypophysis.
The autonomic nervous system (of which I know practically nothing) is part
of the brain's way of controlling the biochemical organism. There is also
autonomic sensing, as discovered relatively recently. As in the pattern
already discussed, the brain, a physically distinct system, directly sets
the reference levels for biochemical systems, and also receives neural and
biochemical information from them. The organ systems consist of specialized
cells which control specific concentrations of biochemical substances, which
serve both as the chemical substrate in which other processes take place, as
as information-carrying signals which set reference signals for lower levels
of organization and which carry information about the results back to the
source to complete the control loop.

Within each cell, there are numerous control processes, but nobody has
looked at these systems to see where their reference signals come from
(although hormones seem one likely candidate). At the level of the cell, the
trail peters out and we must extrapolate to still lower levels. There are
spotty reports of negative feedback systems at the intracellular level, all
the way down to systems involving DNA or RNA. I am mostly lost in this world
of biochemistry, but others like Bob Franza are quietly pursing the matter
from the PCT point of view.

Valentin F. Turchin, a Russian cyberneticist, sees evolution as a series of
"metasystem transitions." A metasystem transition is, in my terms, the
addition of a new level of control to a collection of otherwise independent
control systems that make up an organism at a given time. Cliff Joslyn can
probably give a better account of Turchin's ideas that I can. These
metasystem transitions, as I see them, consist not just of adding more
complex interactions among the physical control systems that exist at a
given stage, but of adding physically distinct cells specialized in a new
way which perform a higher level of control by sending reference signals
directly into the systems that already exist, and receiving copies of the
signals that the existing systems control. Before the new cells come into
existence, the highest reference signals for the existing cells are fixed,
or are determined by natural selection. Where I say "cells," of course, one
could say, at a lower level of organization, "biochemical systems."

The upshot of this is that an organism, whether it consist of one cell or
billions, is a unit unto itself. Its organization, I propose, is that of a
physical hierarchy of control systems, with some organisms having more
levels than others but with no organism directly receiving reference signals
from outside its physical boundaries. The organism, taken as a whole, is an
autonomous unit as far as I can see. It lives or dies by its own efforts.

COULD there be higher levels of control, invisible and inconceivable to us
and of which we know nothing? Of course. Anything we can imagine COULD be.
The range of things that COULD be, if unfettered by demands for evidence, is
infinite. But I am interested only in what IS -- that is, in what there
seems to be some evidence for.

Best,

Bill P.