Bees; Lobsters;Controlling behavior

[From Bill Powers (950524.0915 MDT)]

Bruce Abbott (950520.1130 EST) --

Good stuff on bees from Nachtigall.

    The bee . . . has particular muscles which can change the position
    of the antennae slightly with respect to the air stream. Then the
    air resistance during flight can no longer bend them by the same
    amount as before but rather, for example, a little less. . . . a
    new set point for the flight velocity is achieved.

I would guess that this system is actually analogous to the stretch
reflex, where the reference signal is converted to a length bias on the
spindle and control brings the sensed length to a match. The output of
the annulospiral ending is an error signal. The small muscles in the
bee's control system would exert a force on the antenna, and flying
speed would be varied to apply an equal and opposite force, cancelling
the bias due to the muscle and bringing the mechanical error to zero.

     Bees and flies have two antennae, one right and one left. It has
     been shown that each of these antennae regulates only the amplitude
     of the wing on its own side of the body. This has possibilities
     for flying in a curved course. If one antenna is cut off, the
     insect always flies around in circles. It is only when both
     antennae are operating together that they permit the bee to fly in
     a straight line and compensate for gusts of wind which might push
     it a little away from this line. This is critically important for
     orientation of the flight between hive and food source, which the
     bee should make as straight as possible.

This isn't quite right: to correct for gusts and direction relative to
terrain would require an inertial or visual perceptual system. I seem to
recall that bees have a gyro system which is made of small otoliths on
the ends of short stiff hairs. The hairs are driven to make the otoliths
oscillate, and they tend to retain the same plane of oscillation as the
body turns, providing angular acceleration data. I would interpret the
differential control of antenna bending simply as a means of controlling
the curvature of the flight path relative to the air. In flight,
velocity is always relative to still air.

     But there is one more point to be made: the bee does have a
     second servo system to control its velocity, involving the two
     large compound eyes. The essential purpose of this system is to
     hold constant the velocity over the ground; the antennae, on the
     other hand, regulate the velocity through the air.

This second servo system would operate by varying the reference signal
for the flight-curvature and -velocity control systems, and it would
maintain the speed and path relative to the visual image of the
surroundings. The lower order systems would operate just like those of
the "people" in the Crowd program.

     . . . But an optical measuring device which fixes points on the
     ground and computes how quickly they move backwards, can do this.
     The highly complicated compound eyes are admirably suited to this
     task. They can recognise the dangerous drifts produced by head,
     tail, or side winds and compensate for the error in the signals of
     the antennal control circuit.

Srinivasan carried this a step further, in determining that the
controlled speed variable is visual streaming. The result isn't just
"compensation" for "dangerous drifts," but continuous control of speed
(and through sensing differential outflow, direction).


Avery Andrews (959521:0515)--

     I wonder if the hidden tracking work people have been doing could
     be extended to investigate what kinds of predicted motion people
     can handle. Straight lines, obviously, what about various curves
     and accelerations?

Good point. Can obviously be investigated.
Rick Marken (950521.0030) --

To Clark McPhail:

     You are saying that it is impossible to control behavior WITHOUT
     knowing the reference signals. This implies that WITH knowledge of
     the reference signals, control of behavior is NOT impossible. Your
     final sentence confirms this interpretation; you say that control
     of behavior is possible, though only with great difficulty. PCT
     shows that control of behavior is not possible AT ALL.

Do you really mean that? If I know where you intend to keep the knot in
the rubber-band experiment, I can move your hand around quite freely and
accurately, as long as I don't cause you any problems. Isn't that
control of another person's behavior? Or are you distinguishing
"behavior" from "action?" I presume you would agree that this is an
example of controlling another person's action (or your perception of

From B:CP:

"If producing that behavior creates no inner conflicts, B will produce
it, and A will have "made" B behave as A wanted. " (B:CP, 1973:261)

     Notice that "made" is in quotation marks. The quotation marks are
     essential to the point of the sentence, which is that it _looks_ to
     A like B's behavior has been controlled, but this appearance is an

The illusion is that there is an S-R connection between A's action and
B's action. That is, calling that "control" is a mistake if you're
thinking of control as an open-loop phenomenon. But if you understand
that A is controlling a perception of B's hand by varying the position
of A's hand, and that B is controlling the knot position affected by the
sum of B's hand position and A's disturbance, then there is no illusion.
If you want the knot to be in a certain position, your hand position is
determined by external forces applied to the knot; you no longer have
any choice of how to move your hand.

The rat is controlling the rate of reinforcement by varying its rate of
behavior, and the experimenter is controlling the rat's rate of behavior
by varying the ratio of reinforcements to bar presses. Both are truly
controlling in the PCT sense.

I agree with you that many experiments that look like the Test are
really not -- but some would require fewer added procedures than others.
It would be useful if you would read Garfinkels "breaching" stuff and
suggest to the sociologists what steps could be added to make the
procedure into the Test.
Bruce Abbott (9505xx) --

Thanks to Rick Marken, I finally have the article by Simmer, Meyrand,
and Moulins, "Dynamic Networks of Neurons" (the lobster article) and
have been through it a few times.

It seems that the main point is that by application of signals from
higher centers, functional units of neurons can be temporarily used as
if they were larger units. Considering HPCT, this does not come as a
great surprise, although it's nice to see the detailed work being done.

It would be nicer if the authors weren't so prone to overinterpretation.
Of course that impression may come from the popularization mode of
presentatation; in the original articles there may be a lot more detail.

One problem I have is with the method of recording data. The traces of
Fig. 3 seem to be a combination of impulse recordings with an electrode
(or amplifier circuit) response that rectifies and semi-smooths the
spikes, so we get a quasi-average-frequency measure superimposed on
spike information. There seems to be some high-frequency cross-talk
between electrodes which may have nothing to do with function. I think
it must be very difficult to get a recording that shows physiologically
meaningful variations in single cells with sufficient bandwidth to make
sure we're seeing what is happening in the cell instead of in the
electronic amplifiers or the surrounding medium. An electrode stuck into
the fluids around a cell isn't exactly a high-impedance probe, and who
knows what aspect of cell function it is picking up?.

According to the text, the recordings were made with electrodes on the
outgoing motor axons, not in the cell-bodies as shown in Fig. 3. These
are described as "extracellular" recordings, which means they are
subject to influences that may have little to do with the cell

The main problem with this article (if not the original studies) is that
only the outputs of the cells are shown, with nothing to indicate the
nature of the connections between them. What we seem to have is a neural
oscillator, with the two neurons combined into a circuit that involves
one cell (PY) firing at rates that vary 90 degrees out of phase with the
firing-rate variatioms in other cells. Mutual inhibition alone will
create a flip-flop effect; there must also be time integration to get an
oscillator. I don't have the impression that the studies were
sufficiently detailed to pick up a time integration effect.

It would be interesting to know what the effect of a bidirectional
electrical connection between two cells is. When either one fires, does
it make the other fire, too? Obviously you can't have impulses going in
both directions literally simultaneously; colliding normal and
antridromic impulses should simply cancel, as the mechanism for
propagation has been depleted immediately behind each impulse. The only
other possibility is that the link acts like a electrical wire, forcing
potentials at both ends to be the same.

As to the "bursting" property that is gated on and off by a "modulatory
interneuron" in Fig. 6, this looks like a gated one-shot. An initiating
electrical spike triggers the one-shot on if the enabling input is
present, after which the one-shot maintains itself on until some
internal integrator reaches the threshold needed to sustain the firing.
Then the one-shot turns off. I would expect that other parameters, like
the frequency during the on period and the duration of the on period
would also be adjustable by external signals.

Almost any neuron will behave something like this if presented with a
large enough input stimulus. If a large jolt of excitatory
neurotransmitter is received, the innards of the cell are biased far
above the threshold of firing, and there will be "after-firings" for a
time depending on how extreme the stimulus was (how many parallel inputs
at the same instant) and on the amount of chemical concentration that is
used up or diffused away with each firing. Some cells (for example,
local negative feedback internuncials leading from a motor neuron's
output back to its cell body) have after-firing rates that fall off
exponentially to zero after each impulse, leading to negative leaky-
integral feedback around the neuron, providing a phase-advance circuit
(in terms of frequency of firing).

Everything depends on the parameters. The connections alone tell us only
one part of the story.

I was struck by the complete absence of any discussion of the role of
sensory signals. The reason may be in the authors' stated theory:

     Animals choose constantly among a wide range of behavioral
     capabilities: walking, running, fighting, courting, and so on. An
     animal's nervous system generates each of these activities by
     turning on a specific network of neurons, which produces a
     characteristic sequence of electrical signals, called motor output,
     that instructs muscles to perform particular movements.

Obviously the authors believe that behavior is simply output, with all
actions, however complex and however related to unpredictable
environmental events, merely being emitted and having fixed effects. So
naturally, they concentrate on output functions. If they don't realize
that walking, running, fighting, and courting require constant
readjustment of outputs in order to maintain the patterns we observe,
they will obviously not realize that generating output doesn't explain
behavior. I should think that they would at least mention the fact that
sensory feedback from food in the esophagus, stomach, and pylorus must
have something to do with the way these functions are used.
As to extending these results to larger units of behavior, there are
some severe problems. If simple oscillators were sufficient to explain
things like walking, maybe the parallel would hold, but they aren't. To
prove this to yourself, just start walking and then slow down and
reverse. You can actually freeze the walking at almost any point,
reverse it, continue in the same direction, or go back and forth between
forward and reverse. Maybe this can be accomplished with an oscillator
circuit that has parameter-setting inputs, but it's pretty hard to
imagine. The simple image of an oscillator that produces regular
"pendular" motions of the legs is obviously insufficient. (I've always
thought that the image of "pendular" movements is singularly
inappropriate, as the pendulums are actually upside-down with the fixed
point in contact with the ground).
Nice observations on fly behavior.
Best to all,

Bill P.