plummeting moths; drooling dogs

[_Avery_Andrews]

[from Avery Andrews (920320)]

(Rick Marken 920319)

just get that damn signal outta here, sayeth the moth's
control system.

Well, that's exactly what I'm conjecturing that maybe is *not* happening,
because the changes in the signal induced by the moths behavior aren't
material to survivability of the moth. So my judgement so far would be
that if it what is going on is avoidance of being eaten via quick
departure from the vicinity of the bat and the accompanying sonar
signal, its control, otherwise it ain't. I'm not committed to its
actually being one way or another - just to clarifying the issues.

You might, of course, be right about the salivation stuff - If it works
the way you say, then I entirely agree that that is control. Suppose
salivation in response to smells is acquired. Then that might be
part of a control system designed to forestall error signals, on
the general basis that smell-of-nice-garbage is a harbinger of
dryness-in-the-mouth, which gets forstalled by some anticipatory
salivation. I am intrigued that this looks like
control on a larger time scale, an S-R hookup on a smaller one.
The case of hardwired salivation-in-response to smell doesn't
look control-ish except perhaps at the level of evolution.

You betcha!! Now think about how much you can "do" without influencing
one or another of your sensory inputs. When you realize that that amount
is almost precisely ZERO then you have entered the PCT zone.

But when these influences have no material relevance for what is being
done, I wouldn't want it to call it control. I'd agree that the role
of S-R hookups in human psychology is pretty minimal, but that does not
equal necessarily nonexistent (I won't repeat my previous suggestions
about possible candidates). *I* think it's should be quite useful
to look at cases of possible S-R hookups with an open mind, & describe
the kinds of facts that would induce one to classify them as one or
the other (as I think we're making progress on). I see no reason why
simple critters shouldn't have a fair number of these things rattling
around in their circuitry.

Avery.Andrews@anu.edu.au

[Human_Factors_Lab]

[From Rick Marken (920320)]

Reply to Avery Andrews (920320):

I said:

just get that damn signal outta here, sayeth the moth's
control system.

and Avery said:

Well, that's exactly what I'm conjecturing that maybe is *not* happening,
because the changes in the signal induced by the moths behavior aren't
material to survivability of the moth. So my judgement so far would be
that if what is going on is avoidance of being eaten via quick
departure from the vicinity of the bat and the accompanying sonar
signal, its control, otherwise it ain't. I'm not committed to its
actually being one way or another - just to clarifying the issues.

By controlling the signal strength the moth generally does survive.

From your perspective it looks like the moth is avoiding being eaten;

that is certainly a way to describe the situation. But the moth probably
knows nothing about that. Suppose that we set up a situation where we
create artficial bat noises and, knowing that the moth controls them by
dropping, we have another predator (instead of a nice beetle keeper)
waiting at the bottom of the fall. Now the moth gets eaten as a result
of controlling its sensory input. Are you saying that this means that the
bat is not controlling in this situation -- because it is not avoiding
being eaten?

I guess I'm not sure what you are trying to get at in the above
paragraph. Are you trying to say that the dropping might still be
a response to the stimulus sound even though the dropping changes
the sensory effect of the sound? The only way this could possibly be
true is if the sensor shuts down as soon as the moth starts to drop.
Otherwise, the sensor output (p) is a continuous function of the bat's
location (d) and the moth's position (o) so that

p = k.1d + k.2o (1)

since the falling (change in position, o) is clearly influenced by
the sensor output (p) we also have

o = k.3p (2)

So the behavior os the moth is characterized by two, simultaneous
equations. Equation 2 is the SR law. I'm saying that you can ignore
equation 1 (which makes output depend on itself) only if you eliminate
the effect of p immediately after it is applied -- this would mean
bringing time into the equation. I suppose the equations would
then be

p = k.1d t<s where s is the start of the fall and
o = k.3p t>s

or something like that. But if you just leave things the way they
actually occur in nature then there is a closed loop -- no matter
what. Equations 1 and 2 apply simultaneously. We also know that behavior
is stable -- that means that the loop gain is <= 0. This means that the
coefficient in one of the equations must be 0 or negative. When we
get the signs of the coeffcients right and solve for o we get:

o = -k d

The output of a stable closed loop system depends on the disturbance
to the sensory input -- NOT on the sensory input itself. It looks like
SR (because d looks like a stimulus) but the moth does not respond to
sensory input (o is not a function of p).

I said:

You betcha!! Now think about how much you can "do" without influencing
one or another of your sensory inputs. When you realize that that amount
is almost precisely ZERO then you have entered the PCT zone.

I see no reason why
simple critters shouldn't have a fair number of these things rattling
around in their circuitry.

I can think of no reason why they should. What could possibly be the
value of having inputs cause outputs that have no effect on the
inputs that caused them -- the outputs just going off into the world. Then
organisms would be truly like computers -- a classic SR device (usually).
You put stuff in at the terminal and that input gets churgled (I made it
up -- bet you know what it means) around and turned into output that
has absolutely no effect on the input -- unless there is a person there
to change inputs based on outputs.

Most organisms are built with sensors all over their bodies -- inside and
out. As I said, I can't think of any way that that body could be moved
without it having some sensory consequence. If you also imagine that
some of those sensory inputs are responsible for causing the body to
move in certain ways then you've got an organism that is locked in a
loop -- completely. It is possible that there are sensory causes of
outputs that are completely protected from any effects of those outputs.
Maybe there could be a connection between sensors on the tip of the
tongue and movement of the big toe. But why would there be many of such
connections? Why do organisms generate outputs anyway -- just to give
observers something to see? No, its to help them maintain THEMSELVES in
some way. So if there are SR connections -- pure; no RS connection --
then the R must be occuring for a reason, right. Like to get the
organism out of danger. I suppose that in a HIGHLY STABLE, DISTURBANCE FREE
environment an organism could survive with some SR connections. The
connection from S to R would also have to be HIGHLY reliable. So maybe
you could have an organism that always makes response R when stimulus
S occurs (and S is protected from the effects of R) and the effect of
R on a variable that really matters to the organism can be counted on to
be the same every time. But this has got to be a very rare arrangement.

I think that what happened is that evolution has created machinery (bodies)
housing sensors that can be used to keep these sensor's outputs where they
should be (as determined by the organism) in the context of the environments
in which these bodies happen to find themselves. But what evolution (and
"behavior" for that matter) is all about is sensory input -- PERCEPTION.
Organisms exist only to keep their perceptions where they should be --
the outputs exist ONLY for the benefit of the senses. Unfortunately,
from an observer's perspective, the process of controlling sensory
inputs looks compellingly like like response to stimulation. Moving
past this illusion is what PCT is all about.

These are GREAT questions Avery -- they really get to the nitty-gritty.
Keep pushing and don't let us get away with anything. I think you are
poking around where all psychologists should be poking around. We (PCTers)
need psychologists to ask us these very tough questions -- because it
get's right to the heart of what most psychologists think about PCT,
and that is -- so what? We have to be able to answer the "so what?"
question and we can't answer it in a convincing way unless people ask
questions (or pose problems) like yours. Thanks.

Hasta Luego

Rick

···

**************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
Internet:marken@aerospace.aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)

[Human_Factors_Lab]

[From Rick Marken (920320b)]

Martin Taylor (920320 14:00) says:

Without wishing to disagree with Rick's conclusions that

Not to worry. I'm used to it. Besides, I just took a brief foray this
week into one of the network newgroups. The tone was so different --
angry, hostile, insulting -- compared to CSGNet it was amazing. A real
eye opener. I learned nothing except never to do that again. And the
amount of real information that gets transmitted here relative to there
is astounding. It's incredible what you can accomplish when the goal is
cooperation rather than victory. Gary -- I second my vote to keep us
from becoming a newgroup. Even at the height of the "Beer bash" we
looked like a church social compared to what goes on in those newsgroups.

So far, we have: Moth senses bat sonar; moth changes mode of operation by
folding wings and dropping; moth reduces sensation of bat sonar. But we
don't have any sense of a feedback loop with measurable gain. We need the
next stage: Moth's sensation of bat sonar in relation to a reference level
leads to some change in error signal that results in some behaviour.

I think not. The equations that I gave in the Avery article have no
explicit reference input -- there is no theory of the organism other
than an SR theory -- o = kp -- but when you solve them simultaneously
you end up with the basic facts of negative feedback control:

o = kd and

p = 0

(because there is no reference, with high gain the sensory input is driven
to the 0 of the perceptual variable scale.)

There are some assumptions about gain that I made -- but they can be
based on inspection. The gain is not 0 because there is output to
input. The gain is negative because, if you get the signs of the
coeficients so that gain is positive then the equations make no sense
as descriptions of behaivor.

Bill P. did all this stuff much more elegantly and beautifully in the
Psych Review article (1978). He basically showed that, with negative
feedback from output to input there IS control -- and an SR model of
the situation is simply WRONG.

Something here seems to me very S-R.

Yes, equation 2 describes the organism as an SR system. The FACT of control
is derived from recognition that there is also an RS law and that
taking both facts into account simultaneously reveals that the system is
CONTROLLING.

No matter what the other circumstances (we
hypothesise) an observer sees a sequence: bat sonar -> moth drop. There
seems to be no disturbance on the bat sonar percept that would or could
affect this (unless it be a bat-mute!).

The disturbance is to the sensory variable that causes the moth drop.
One major disturbance to this variable is movements of the bat -- think
of my d as the effective intensity of the sound, which depends on how far
the bat is from the moth at any instant. As the bat moves, d changes
and, hence, p changes.

             there seems no way to discriminate between S-R and
control.

There is -- but it's not simple (obviously -- or everyone would have
noticed by now).

If you have a predisposition to see all behaviour as control, you can easily
describe the moth's behaviour as control. If you want to believe that some
behaviour is S-R driven, you can see the moth's actions that way, and I do
not believe that anything other than Occam's razor will say that you are
wrong to do so.

Control can be tested -- and proved to exist. The difference between SR
and control is BIG and needs no razor for choice. It just requires
studying the equations, doing the tests and looking at what is going on.
I think the continued predilection for seeing behavior in SR terms is
purely psychological -- it's easier and more familiar and it takes little
(or no) learning to understand it. I think eventually control will be
accepted -- just as the fact that the earth spins on its axis is now
accepted. All the obvious evidence is against both. I think people eventually
accepted the spining earth just because all the experts said so (of course,
when we could take pictures of it that made it even easier). It didn't
take occam's razor to make Copernicus right -- it took tough, detailed
thinking. Most people will not want to do this -- so when control theory
is finally accepted in the life sciences, people will just accept is
unquestioningly (as they accept a spining earth) and wonder how people
could have been so silly as to believe in SR -- they will even forget that
the SR view (like the flat stationary earth view) is the one that is by
far the more obvious.

Hasta Luego

Rick

···

**************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
Internet:marken@aerospace.aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)

[_Martin_Taylor]

[Martin Taylor 920320 14:00]
(Rick Marken 920320)

Without wishing to disagree with Rick's conclusions that
"Organisms exist only to keep their perceptions where they should be --
the outputs exist ONLY for the benefit of the senses", I do think that his
argument is incomplete in respect of the dropping moth.

So far, we have: Moth senses bat sonar; moth changes mode of operation by
folding wings and dropping; moth reduces sensation of bat sonar. But we
don't have any sense of a feedback loop with measurable gain. We need the
next stage: Moth's sensation of bat sonar in relation to a reference level
leads to some change in error signal that results in some behaviour. Something
here seems to me very S-R. No matter what the other circumstances (we
hypothesise) an observer sees a sequence: bat sonar -> moth drop. There
seems to be no disturbance on the bat sonar percept that would or could
affect this (unless it be a bat-mute!). Even if we take it that the internal
structure is an ECS that has as a reference signal zero-bat-sonar, the
effect is still S-R-like. There would be more apparent control if the
reference for bat sonar were non-zero, so that the moth sought out bats
if it didn't get enough, but even in that case, a bat could CONTROL the
moth if the reference level for bat sonar were fixed. It would still look
like S-R for all practical purposes. And when the fixed reference level is
zero (presumably plus some small increment to allow for low-signal detection
problems) and the behaviour is always the same cessation of flying (or drop
off a resting place?), there seems no way to discriminate between S-R and
control.

The fact that the moth recommences flying when the sonar signal goes away
can equally well be seen as S-R: Signal->response (stop flying and drop); no
signal->no response. There is or isn't a feedback loop, depending on how
you look at it. Certainly the moth's behaviour changes its environment,
altering the stimuli to which an S-R person would say it responds. Certainly
the moth's control system brings its percepts closer to their reference
states. Who wins here?

If you have a predisposition to see all behaviour as control, you can easily
describe the moth's behaviour as control. If you want to believe that some
behaviour is S-R driven, you can see the moth's actions that way, and I do
not believe that anything other than Occam's razor will say that you are
wrong to do so. Will Occam's razor actually shave away the S-R interpretation?
I don't know, but I have faith that it will.

Martin

[_Martin_Taylor]

[Martin Taylor 920320 18:40]
(Rick Marken 920320b)
I didn't follow your post to Avery on the moth feedback loop, and I don't
follow your rebuttal to my claim that one could see the moth's behaviour
in S-R terms. Here's the quote and problem.

Are you trying to say that the dropping might still be
a response to the stimulus sound even though the dropping changes
the sensory effect of the sound? The only way this could possibly be
true is if the sensor shuts down as soon as the moth starts to drop.
Otherwise, the sensor output (p) is a continuous function of the bat's
location (d) and the moth's position (o) so that

p = k.1d + k.2o (1)

since the falling (change in position, o) is clearly influenced by
the sensor output (p) we also have

o = k.3p (2)

So the behavior os the moth is characterized by two, simultaneous
equations. Equation 2 is the SR law. I'm saying that you can ignore
equation 1 (which makes output depend on itself) only if you eliminate
the effect of p immediately after it is applied -- this would mean
bringing time into the equation. I suppose the equations would
then be

p = k.1d t<s where s is the start of the fall and
o = k.3p t>s

or something like that. But if you just leave things the way they
actually occur in nature then there is a closed loop -- no matter
what. Equations 1 and 2 apply simultaneously. We also know that behavior
is stable -- that means that the loop gain is <= 0. This means that the
coefficient in one of the equations must be 0 or negative. When we
get the signs of the coeffcients right and solve for o we get:

o = -k d

The output of a stable closed loop system depends on the disturbance
to the sensory input -- NOT on the sensory input itself. It looks like
SR (because d looks like a stimulus) but the moth does not respond to
sensory input (o is not a function of p).

In (1) you have the sensor output as being the sum of two linear functions
of the respective locations, whereas I think it should be a non-linear
function of the vectorial difference of the two positions with terms in
the bat's orientation relative to the moth. The appropriate form should
probably be something like (ignoring the bat orientation)

p = f(o-d) (1a)

and (2) should be in terms of the derivative of o, not of o itself (actually
the derivative of the z component of o would be closer, but let's not worry
about that). So I think we should have

do/dt = g(p) (2a)

The moth "wants" p to be minimized. These equations in themselves give no
hint as to how that can be done, so far as I can see, since (2a) is really,
as you say, a statement of the S-R behaviour, and whether dropping will
increase or decrease p depends on whether the moth is above or below the
bat's beam.

The equations cannot be solved, because integrating (2a) provides a constant
of integration that depends on the initial value of g(p). g(p) itself is
a problem, because it probably is discontinuous, switching between 0 and
some fixed value at some small value of p. All we can get, so far as I
can see, is

do/dt = g(f(o-d)) (3a)

which is a statement of the S-R "event" in externally observable variables.

I don't see any demonstration of control here, still. I see consistency
with control, and consistency with an S-R view. The only reason, still,
that I see for preferring the control view is that it is more parsimonious,
and covers a wide variety of situations including this, whereas the S-R
view can be applied only under special circumstances, all (?) of which
can be seen as degenerate instances of control.

As Bill has often said, if the world is stable, it is possible for the same
actions to have the same effects most of the time, and you don't NEED control,
even if the opportunity for control is inherent in the situation. Whether
control is actually exercised when the opportunity is there is a matter for
experiment.

I suppose that moths that drop into the bat's path rather than out of it
get eaten, but on average moths that drop get eaten less often than moths
that ignore the bat. They don't know it, as Bill says, but Nature knows
whether they have offspring.

Martin