Self-coercion

[From Bruce Abbott (980526.1335 EST)]

Rick Marken (980524.1120) --

It may be worth noting that this seems to apply to self-coercion
as well.

Self-coercion? So there must be two independent control systems in the same
person, both attempting to control the same variable but at different
references, one of which is capable of exerting overwhelming force relative
to the other, thus "having its way" with the controlled variable. Why
doesn't a higher-level system simply set the gain of the to-be-coerced
system to zero and eliminate the conflict? Wouldn't that be easier?

Maybe this is why abstinance sometimes helps solve
addictions; you prevent yourself from getting the perception you
want (by keeping yourself away from the Jack Daniels bottle, say);
the error keeps growing until it falls into the regime of the
"universal error curve" -- the increased discrepency between
the desired and actual perception of the amount of Jack Daniels
consumed leads to a _decrease_ rather than a further increase in
the error signal. The system controlling for the amount of Jack
Daniels consumed seems to "give up". Of course, it quickly comes
back into operation if you sudddenly have easy access to a shot
of JD.

Extremely unlikely, in my estimation. From what I understand of the matter,
the extreme craving diminishes with sufficient time (after rapidly reaching
a peak), so the error would be decreasing, not increasing, from that point
on. Other factors are also at work involving higher levels of control --
the recognition that the long-term consequences of continuing to drink are
destroying one's ability to achieve and maintain other important reference
states (like self-respect, the love and respect of family and friends,
ability to earn a living, and so on), leading to control actions to minimize
temptation, strengthen social support, and so on).

Regards,

Bruce

[From Rick Marken (980526.1540)]

Bruce Abbott (980526.1335 EST) --

Self-coercion? So there must be two independent control systems
in the same person, both attempting to control the same variable
but at different references, one of which is capable of exerting
overwhelming force relative to the other

Yes. I think it's also called "self-control, "will power" and
"white knuckling it".

Why doesn't a higher-level system simply set the gain of the
to-be-coerced system to zero and eliminate the conflict?

What would the higher level system be controlling by doing this?
Are you proposing an all-purpose conflict resolution system that
resolves conflict by reducing the gain of one of the systems in
conflict? If so, where is this system located in the control
hierarchy? Or are you imagining many conflict-resolution specific
control systems distributed throughout the control hierarchy? How
does this system (or systems) know _which_ of the conflicted
systems to lower the gain on?

Wouldn't that be easier?

It's easy to think about but I think there are many complications
(such as those I mention above) that come up when you actually try
to _model_ the process. Another complication (besides those
mentioned above) would come up once you've got the higher level
gain reduction system working: the problem is that the now weaker,
low gain system will have a hugh error signal as long as the higher
gain system is winning the conflict. This error might lead to
unpleasant side effects in the real system, such as the constant
production of high levels of adrenalin.

I hope you're planning to attent the CSG meeting in Vancouver to
present some of your PCT research and modeling results.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Abbott (980526.2030 EST)]

Rick Marken (980526.1540) --

Bruce Abbott (980526.1335 EST)

Self-coercion? So there must be two independent control systems
in the same person, both attempting to control the same variable
but at different references, one of which is capable of exerting
overwhelming force relative to the other

Yes. I think it's also called "self-control, "will power" and
"white knuckling it".

Yes.

Why doesn't a higher-level system simply set the gain of the
to-be-coerced system to zero and eliminate the conflict?

What would the higher level system be controlling by doing this?

Higher-level variables like self-respect.

Are you proposing an all-purpose conflict resolution system that
resolves conflict by reducing the gain of one of the systems in
conflict?

No. The idea of turning down the gain in order to stop controlling a
variable has been proposed before as a possible way a control system may be
"turned off."

The gain idea is not crucial. More in line with HPCT, why couldn't the
higher-level system simply set the lower-level system's reference to zero?
That would have two positive effects: (1) eliminate error in the lower-level
system and (2) allow the higher-level system to reduce its own persistent
error by eliminating the conflict at the lower level.

It would seem that this does not happen in the example you gave. If HPCT
permits this solution, then the interesting question is why doesn't it happen?

Wouldn't that be easier?

It's easy to think about but I think there are many complications
(such as those I mention above) that come up when you actually try
to _model_ the process. Another complication (besides those
mentioned above) would come up once you've got the higher level
gain reduction system working: the problem is that the now weaker,
low gain system will have a hugh error signal as long as the higher
gain system is winning the conflict. This error might lead to
unpleasant side effects in the real system, such as the constant
production of high levels of adrenalin.

Not if the gain has been set to zero. If the gain is zero, no amount of
error will yield any production of adrenalin. e*0 = 0

I hope you're planning to attent the CSG meeting in Vancouver to
present some of your PCT research and modeling results.

Alas, Vancouver is a bit too distant for my budget. You guys ought to
consider having a CSG meeting out east once in a while.

Regards,

Bruce

[From Rick Marken (980527.0850)]

Bruce Abbott (980526.2030 EST) --

why couldn't the higher-level system simply set the lower-level
system's reference to zero? That would have two positive effects:
(1) eliminate error in the lower-level system and (2) allow the
higher-level system to reduce its own persistent error by
eliminating the conflict at the lower level.

I think you really need to look at a model of conflict and see how
it works. A conflict exists because higher level systems require the
same lower level perception to be at two different reference levels
simultaneously.

What higher level system are you talking about? If it's one of the
systems that is setting incompatible references for the lower level
("conflicted") perception then it can't arbitrarily just set the
reference for this lower level perception to zero; it is trying to
set the lower level perception to a particular reference value as
the means of controlling it's own perception.

If this higher level system is a system that set's the reference for
one of the systems that is _creating_ the conflict, then setting the
reference of the lower level, conflict creating system to zero
would, indeed, solve the conflict. But this higher level system is
setting the reference for the conflict creating system to something
other than zero in order to control _its own_ perception; this higher
level system can't "simply" (arbitrarily) set the lower level, conflict
creating system's reference to zero; this higher level system can
only set the lower level system's reference to a value proportional
to the error (or to the integral of the error) in the higher level
system itself.

Moreover, this higher level system probably sends its output to many
lower level systems simultaneously (see my spreadsheet), only one of
which is the conflict creating system. So even if this higher level
system could arbitrarily set it's output to zero, that would solve
the conflict but it would also cause all the other lower level
systems to bring their perceptions to the _wrong_ level for
controlling the higher level system's perception; the higher level
system would win the conflict "battle" but lose the control of
perception "war".

You could see how all this works by studying the behavior of my
spreadsheet hierarchy. It's easy to create conflicts in the
hierarchy. Just have two level 3 or level 2 systems control the
same level 2 or level 1 perceptions, respectively. Then see if you
can design the hierarchy to "solve it's own conflict". I think
you'll see the problem then.

If HPCT permits this solution, then the interesting question is why
doesn't it happen?

HPCT doesn't "permit" the solution you describe because it is not
a solution. Conflicts can only be solved by some capability that is
not part of the HPCT hierarchy. That capability is called
"reorganization" and it probably involves consciousness. We don't
fully understand this process because there has been very little
research on it.

If the gain is zero, no amount of error will yield any production
of adrenalin. e*0 = 0

The gain of the adrenalin production control loop is not zero just
because the gain of the perceptual control system is zero.

Alas, Vancouver is a bit too distant for my budget.

For your budget or for your system concepts? :wink:

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Nevin (980527.1215 EDT)]

Rick Marken (980527.0850)--

I'm getting confused. If a reference-level signal is set to zero, does that
not mean that the comparator produces an error signal if the perceptual
input is anything other than zero? I'm sure this has been clarified before.
Bear with me, I'm a PCT student of little brain.

  Bruce Nevin

[From Bill Powers (980527.0845 MDT)]

Bruce Abbott (980526.2030 EST)--

Why doesn't a higher-level system simply set the gain of the
to-be-coerced system to zero and eliminate the conflict?

Any conflicts that are that easy to eliminate don't last long enough to
matter.

The hard conflicts arise because _both_ sides are important to higher-level
systems. I'm sure you can think of examples.

Best,

Bill P.

[From Bruce Abbott (980527.1210 EST)]

Bruce Nevin (980527.1215 EDT) --

I'm getting confused. If a reference-level signal is set to zero, does that
not mean that the comparator produces an error signal if the perceptual
input is anything other than zero? I'm sure this has been clarified before.
Bear with me, I'm a PCT student of little brain.

I think you're confusing yourself with Pooh. Be that as it may, don't
forget that biological control systems are mostly (entirely?) one-way
systems; the error signal can only go positive. The comparator will indeed
produce an error signal if the perceptual input is _above_ zero. In the
example we've been using, setting the reference for imbibing Jack Daniels to
zero would result in error only if some other control system were "forcing"
one to drink. Absent that, there would be no motivation to drink. Of
course, the evidence shows that the alcoholic is not able simply to reset
his or her reference for having a drink (or more accurately, the perceptual
consequences of being drunk) to zero, and therein lies the problem when some
other system in that same person wants to avoid some of the other
consequences of being drunk, by refusing to take that drink.

Bruce

[From Bruce Nevin (980527.1413 EDT)]

Bruce Abbott (980527.1210 EST) --

biological control systems are mostly (entirely?) one-way
systems; the error signal can only go positive.

I thought that the reference input is assumed to be excitatory (positive
sense) and the perceptual input to be inhibitory (negative sense); and that
the error output can be either. I got this notion from reading e.g. B:CP p.
62, the challenge to the reader to design a comparator that generates "one
output signal for positive errors and another for negative errors. (This is
necessary because neural currents cannot change sign.)" (The description of
a subtractor is on p. 28 in Chapter 3.)

The comparator will indeed
produce an error signal if the perceptual input is _above_ zero.

By the above, the perceptual input is always below zero (inhibitory to some
amount) or zero (no firing of the nerve).

Suppose

  r = 5 spikes per unit of time (excitatory)
  p = 5 spikes per unit of time (inhibitory)

then

  e = o spikes per unit of time (neither excitatory nor inhibitory)

But suppose

  r = 5 (excitatory)
  p = -10 (that is, 10 inhibitory)

Then

  e = -5 (that is, 5 inhibitory)

Just so, if

  r = 0
  p = -5

then

  e = -5

Have I got this wrong?

  Bruce Nevin

···

In the
example we've been using, setting the reference for imbibing Jack Daniels to
zero would result in error only if some other control system were "forcing"
one to drink. Absent that, there would be no motivation to drink. Of
course, the evidence shows that the alcoholic is not able simply to reset
his or her reference for having a drink (or more accurately, the perceptual
consequences of being drunk) to zero, and therein lies the problem when some
other system in that same person wants to avoid some of the other
consequences of being drunk, by refusing to take that drink.

Bruce

[From Bill Powers (980527.1317 MDT)]

Bruce Nevin (980527.1413 EDT)]

Bruce Abbott (980527.1210 EST) --

biological control systems are mostly (entirely?) one-way
systems; the error signal can only go positive.

I thought that the reference input is assumed to be excitatory (positive
sense) and the perceptual input to be inhibitory (negative sense); and that
the error output can be either.

It is possible for reference signals to be excitatory or inhibitory;
perceptual signals would then be the opposite, for a given control system.
When reference signals are excitatory, setting them to zero turns the
one-way control system completely off: inhibitory perceptual signals alone
can't produce any error signals, or any output. Work it out:

          > Reference signal
          > excitatory inhibitory
    Perc | zero nonzero zero nonzero
   Signal |--------------------------------------------
          >
   zero |
          >
          > [error signal]
  nonzero |
          >
          >

Best,

Bill P.

[From Bruce Abbott (980527.1455 EST)]

Bill Powers (980527.1317 MDT)]

It is possible for reference signals to be excitatory or inhibitory;
perceptual signals would then be the opposite, for a given control system.
When reference signals are excitatory, setting them to zero turns the
one-way control system completely off: inhibitory perceptual signals alone
can't produce any error signals, or any output.

And in that case, the magnitude of the error signal is limited by the
magnitude of r. For example, if e = k(r - p) and p = 0, then e = kr.

Regards,

Bruce

[From Bruce Abbott (980527.1615 EST)]

Bruce Nevin(980527.earlier) --

Bruce, I don't have access at the moment to your post regarding the
computation of error signals (it's now stored locally on my computer at
work), but I thought I'd go over the logic with you anyway.

It's important to keep in mind that the various nervous-system signals are
presumed to be represented as a frequency of neural impulses (which cannot
fall below zero), whereas the interaction of these signals takes place via
release of neurotransmitter at the synapse, which can have either an
excitatory or inhibitory effect. [More complex effects are possible, but
let's keep it simple here.] Consequently, even though a neuron's output
frequency may be zero, the inhibitory input may be so strong that it would
take considerable excitatory input to overcome the inhibition and produce a
nonzero output frequency. In other words, the representation within the
neuron of the net effects of excitation and inhibition can range from
extreme negative, through zero, to extreme positive, even though the output
frequency can only be either zero or positive. Moveover, it will take a
stronger excitatory signal to move the output frequency above zero, as the
inhibitory signal becomes stronger. Once the neuron's output is completely
inhibited, further inhibition will shift upward the rate of excitatory input
required to produce non-zero output, but it will not generate a negative
output frequency.

Regards,

Bruce

[From Bruce Nevin (980527.1754 EDT)]

Bill Powers (980527.1317 MDT)--

When reference signals are excitatory, setting them to zero turns the
one-way control system completely off: inhibitory perceptual signals alone
can't produce any error signals, or any output.

Got it. Thanks. (Nothing to work out, it's obvious.)

It is possible for reference signals to be excitatory or inhibitory;
perceptual signals would then be the opposite, for a given control system.

If a given control system is set up the opposite way, with perceptual input
pexcitatory, then setting reference input to zero would have the effect of
passing p directly on to error output e, wouldn't it? Zero inhibition of
the current perceptual input. I think I'm remembering now that this is a
reason for the assumption that r is excitatory and p is inhibitory. (Not
coercion, but not control either.)

  Bruce Nevin.

[From Bruce Nevin(980527.1845 EDT)]

Bruce Abbott (980527.1615 EST)--

Thanks for the clarification. This is what I get.

Subtractor with p inhibitory and r excitatory:

p r e
10 0 0
10 1 0
. . .
. . .
. . .
10 8 0
10 9 0
10 10 0
9 10 1
8 10 2
. . .
. . .
. . .
0 10 10

Subtractor with p excitatory and r inhibitory (contrary to what I recently
said):

p r e
10 0 0
10 1 9
. . .
. . .
. . .
10 8 2
10 9 1
10 10 0
9 10 0
8 10 0
. . .
. . .
. . .
0 10 0

So a pair of subtractors can operate in tandem so as to yield two error
signals, one in each sense. As the values of p and r vary, both come to e=0
at the same point, but while either error signal has a non-zero value the
other stays at zero.

The effect of "excess" inhibition cannot be reflected in the error output
signal that results from it (which cannot go below zero) but is reflected
in the value of the other error signal, which has the opposite sense.

  Bruce Nevin

[From Bruce Abbott (980527.2140 EST)]

Bruce Nevin(980527.1845 EDT) --

Bruce Abbott (980527.1615 EST)

Thanks for the clarification. This is what I get.

. . .

So a pair of subtractors can operate in tandem so as to yield two error
signals, one in each sense. As the values of p and r vary, both come to e=0
at the same point, but while either error signal has a non-zero value the
other stays at zero.

The effect of "excess" inhibition cannot be reflected in the error output
signal that results from it (which cannot go below zero) but is reflected
in the value of the other error signal, which has the opposite sense.

By Jove, I think you've GOT it! (:->

Bruce

[From Bill Powers (980528.0243 MDT)]

Bruce Nevin(980527.1845 EDT)--

I love seeing you guys playing with numbers. This is how you really learn
about control systems.

The effect of "excess" inhibition cannot be reflected in the error output
signal that results from it (which cannot go below zero) but is reflected
in the value of the other error signal, which has the opposite sense.

Excellent. That is one very acceptable model of how to make two-way control
out of two one-way systems. Here's another consideration.

Suppose the range of the perceptual and reference signals is from 0 to 10.
And suppose the range of the output is also 0 to 10 (you can always scale
the measurement units to make this true). If the gain in the output
function is 10, this means that the maximum meaningful error signal is 1
unit (any greater error will just saturate the output function).

What this implies is that normal control can occur in the range of
reference signals between 1 and 9 units, with some loss of control outside
that range. This is independent of which way the signs are arranged for
perceptual and reference inputs to the comparator. For the inhibitory
perceptual signal, some external disturbance (like gravity or an opposing
control system) is needed.

Best,

Bill P.

[From Bill Powers (980528.0214 MDT)]

Bruce Nevin (980527.1754 EDT)--

It is possible for reference signals to be excitatory or inhibitory;
perceptual signals would then be the opposite, for a given control system.

If a given control system is set up the opposite way, with perceptual input
excitatory, then setting reference input to zero would have the effect of
passing p directly on to error output e, wouldn't it? Zero inhibition of
the current perceptual input. I think I'm remembering now that this is a
reason for the assumption that r is excitatory and p is inhibitory. (Not
coercion, but not control either.)

Setting a reference signal to zero tells the control system to act so as to
keep the perceptual signal at zero (in other words, avoid experiencing it).
If the reference signal is excitatory, we have to interpret "zero" to mean
"very small." Otherwise presence of an unwanted inhibitory perceptual
signal could not lead to any behavior. When the perceptual signal is
excitatory, a zero reference signal means, as you say, that any perceptual
signal at all will generate an error signal. With this arrangement it is
possible to act to keep the perception very close to zero -- the greater
the amplification in the output function, the closer to zero the perception
can be controlled.

Raising the inhibitoryt reference signal above zero means that a larger
perceptual signal has to occur before action occurs. So control still
occurs, but it requires an external bias that tends to make the perceptual
signal very large, with the action then working to reduce the perceptual
signal. An example is temperature regulation in the region where sweating
occurs. Sweating reduces an excessive body temperature, but of course does
not occur for body temperatures below the set point. When you "have a
fever," the temperature reference signal rises. As a result you feel cold
and start doing things to raise your temperature -- get under blankets,
shiver, turn up the thermostat. Only when your body temperature rises above
the new reference temperature do you start doing things to cool off: fling
back the covers and break into a sweat.

Many control systems actually work in pairs, one system's output acting
oppositely to the other system's output, and both being one-way systems.
The systems for flexing the arm at the elbow are an example; the opposed
pairs of eye muscles are another. In the spinal cord, an excitatory
reference signal entering the motor-neuron comparator for one muscle system
also, via an internuncial neuron, inhibits (has a negative-going effect on)
the motor neuron of the opposing control system. Reference signals entering
opposing motor neurons from higher systems, it stands to reason, would
occur in push-pull pairs: if one signal increases the other decreases.
Increasing and decreasing both signals at the same time would affect muscle
tone without generating any net torque around a joint.

The reason for assuming a positive reference signal and negative perceptual
signal in our diagrams is only that we need to use _some_ convention in our
standard diagram. Usually we assume a two-way system, so one direction
_has_ to be positive! Obviously, if the output can act in both directions
around zero, it just about has to be a composite of two output functions in
two different one-way control systems -- muscles can't push.

Best,

Bill P.

[From Bill Powers (980528.0208 MDT)]

Bruce Abbott (980527.1455 EST)--

It is possible for reference signals to be excitatory or inhibitory;

...

And in that case, the magnitude of the error signal is limited by the
magnitude of r. For example, if e = k(r - p) and p = 0, then e = kr.

True. But if the gain of the output function is reasonable -- say, 10 to
100 -- the maximum meaningful error signal will be a small fraction of the
range of the reference signal, so for all reference signals greater than 1
- 10 percent of the range, the lower limit doesn't come into play.

Best,

Bill P.

[From Rick Marken (980528.0810)]

Bruce Abbott (980527.1455 EST) --

And in that case, the magnitude of the error signal is limited by the
magnitude of r. For example, if e = k(r - p) and p = 0, then e = kr.

Bill Powers (980528.0208 MDT) --

True. But if the gain of the output function is reasonable -- say,
10 to 100 -- the maximum meaningful error signal will be a small
fraction of the range of the reference signal

Nice point. Nice illustration of the difference between open and
closed loop thinking.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Abbott (980528.1430 EST)]

Rick Marken (980528.0810) --

Bruce Abbott (980527.1455 EST)

And in that case, the magnitude of the error signal is limited by the
magnitude of r. For example, if e = k(r - p) and p = 0, then e = kr.

Bill Powers (980528.0208 MDT)

True. But if the gain of the output function is reasonable -- say,
10 to 100 -- the maximum meaningful error signal will be a small
fraction of the range of the reference signal

Nice point. Nice illustration of the difference between open and
closed loop thinking.

It has nothing to do with open- versus closed-loop thinking. The fact that
error is usually kept small when the control system is doing its job is an
important point, but it shouldn't keep one from considering what happens to
the system when control begins to fail. That is, after all, what we've been
considering when discussing coercion.

If the perceptual signal is negative, then the maximum error is limited by
the magnitude of the reference signal. If the system is being overwhelmed
(coercion), error cannot grow without limit; in fact it can never exceed the
magnitude of the reference level. This has potential implications for
Bill's concept of a "universal error curve," which is based on the idea that
error continues to grow until it enters a region in which further increases
yield decreasing output.

Regards,

Bruce

[From Rick Marken (980528.1320)]

Bruce Abbott (980528.1430 EST) --

It has nothing to do with open- versus closed-loop thinking.

Right :wink:

If the perceptual signal is negative, then the maximum error is
limited by the magnitude of the reference signal.

Only in one direction.

If the system is being overwhelmed (coercion), error cannot grow
without limit; in fact it can never exceed the magnitude of the
reference level.

On one side, going one way.

This has potential implications for Bill's concept of a "universal
error curve,"

No, I'm afraid it doesn't. To test the idea of a universal error
curve you have to do experiments -- _control system_ experiments,
not the kind you teach your students in your Research Methods
class -- to see what actually happens when you force a controlled
variable away from it's reference level.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken