Some of the things being said about loops now are getting to be a bit more
like some of the ideas I was trying to push a few months ago (`e-control'),
so I'll try another foray in that general direction.
DNA variants do better or worse due to their differing abilities to maintain
conditions suitable for their replication, so what is fundamental is
what I'll call `condition maintainence'. Usually, the best way to
attain this is to have a perceptual function P that approximately
perceives the relevant condition C (exact 100% reliable perception
tends to be too expensive, if it's even possible), and a feedback loop
controlling the value of P, and thereby, indirectly, maintaining C (which will
be more or less tightly coupled to P). So control systems typically
get selected for on the basis of their ability to maintain conditions,
which either contribute directly to DNA replication (Darwinian selection),
or to the control of perceptions of superordinate systems (selection by
(some kind of) reorganization), which are themselves being selected for
on the basis of their ability to maintain conditions.
My picture would be something like this:
···
-------------------------
I | Critter |
--> ---------->|----- |
^ | | |
> > P v Ref |
-->C | ECS<-------- |
^ | | |
> O | | |
<----------|<---- |
> >
-------------------------
What gets controlled by the ECS is some function of I, designated by
P, but what the system is being selected for doing is maintaining
some circumstance C. But due to possibility of disturbances in between
C and I, or limitations on available perceptual and computational
resources, what actually gets controlled is not always exactly
what ought to be (though in systems with a good reorganization
facility and plenty of perceptual and computational resources,
P will track C *very* closely). Nota Bene: C is not something that
the organism is aware of, but is rather the reason (as determined
by our analysis of the whole situation) for the survival or otherwise
of the organism, or the success or failure at satisfying some
superordiante goal).
To get a useful stabilization effect, the C->C loop gain through
the organism has to be significant w.r.t. the disturbances effecting
C. In principle, C->C loop gains could achieve oscillations or
chaos, but is there ever selection for this happening in a loop
going through the environment? At any rate, I'd want to say that
a `control system' was an organism internal setup that has
been selected (via evolution or reorganization) for because of
its ability to stabilize a condition C via negative loop gain.
Atlhough I->I loop gains presumably always exist, it seems to me that
C->C ones needn't. For example, C might be `be undetected by bats'.
C is *not perceived* by moths. Instead (perhaps), moths perceive
evidence of the presence of a hunting bat, and if they can't manage
to zero that perception, do something else, which makes them look
like a falling leaf. So C, which is important for survival, manages
to get maintained without being perceived.
Condition Maintenance via control and S-R chaining does not look very
exhausitive to me--I wonder what the other possibilities are?
Avery.Andrews@anu.edu.au
w