[From Bill Powers (9701208.1730 MST)]
Tracy Harms (1997;01,08.20.00 UTC) --
I've been thinking about various prominent things which are clearly
manipulated, but which are not controlled. As an example, consider the
restfulness of a sleeping chamber when a person retires. If the area is
not sufficiently restful, action will be taken to increase that factor.
However, if restfulness of the room goes above the degree required, *no
action is taken to reduce it*. Therefore we easily conclude that it is
not controlled in the PCT meaning of the term.
This is known in PCT as "one-way control." Actually, when you consider how
neural comparators must work, all neural control has to be accomplished by
one-way control: neural signals can signifiy either a negative or a positive
quantity (like cold versus heat), but the signals must be carried in
separate channels because they can't change sign. You can't have a neural
frequency less than zero.
For convenience we usually model behavior with bidirectional control systems
in which signals can go freely across zero, between positive and negative.
However, the same models could also be expressed as a pair of systems, one
handling negative errors (using positive signals) and the other positive
errors, with the output driving lower-level reference signals (always
positive) so lower-level error signals always signify (positive) actions,
although in opposite physical directions.
Once nice feature of one-way systems acting in pairs is that when you set
both the positive and the negative-meaning reference signals to zero, the
composite control system is simply turned off: it will respond neither to
positive nor negative errors. This allows higher-level systems effectively
to select and deselect lower systems, with systems that are receiving
neither positive nor negative (-signifying) reference signals really being
turned off, rather than controlling for a reference level of zero input. The
drawback is that you can't control for _exactly_ zero input; if you want the
compositive system to work, you have to give it a low but not zero reference
level. But I can think of many cases where that seems realistic.
The above is all based on control systems in which the reference signal is
positive (excitatory) and the perceptual signal is negative (inhibitory in
its effects, although still being a signal with a positive, nonzero,
frequency). It's possible to have comparators with the perceptual and
reference signals having interchanged effects: an inhibitory reference
signal and an excitatory perceptual signal. There is evidence, both
behavioral and neuroanatomical, that this occurs in the brain stem, where
the signals from higher systems seem to enter motor nuclei with inhibitory
effects (through an internuncial cell), while sensory feedback signals arise
via collaterals and arrive with the excitatory sense. A cat with its brain
sliced off just above the brain stem (cutting off signals from higher
systems) exhibits "decerebrate rigidity" with its limbs fully and actively
extended. Apparently the significance of a zero perceptual signal at that
level corresponds to full extension. Any degree of perceptual signal is an
error when the inhibitory reference signal is missing, so the control
systems act to bring the perceptual signal to zero -- corresponding to full
extension of the limbs.
Inhibition is apparently always brought about by an interposed internuncial
cell, a Renshaw cell, specialized to produce inhibitory neurotransmitters.
As far as I know, no cell has an axon that produces both inhibitory and
excitatory neurotransmitters at its various terminal branches.
So, Tracy, you have detected a real phenomenon here. If we ever get PCT to
the point where we don't spend all our time fending off opposition and
steering it back to the main track when people want to run with it in all
directions but the direction of progress in PCT, maybe we can explore such
niceties more thoroughly.
Best,
Bill P.