Side Effects

[From Rick Marken (991019.1100)]

Bruce Nevin (991019.1042 EDT)--

A loop controlling a tactile pressure intensity perception
could possibly have many side effects if a finger is pushing
a button or a key on a keyboard, but this is exceptional.

All control loops produce many (probably an infinite number
of) side effects. Your example of a system controlling the amount
of groceries in the kitchen will produce a zillion side effects;
these side effects are effects (on _any_ variables) produced by
the actions the system takes to maintain the amount groceries
at the desired level. So, as this system acts to control the
amount of groceries, it will affect variables like 1) your
distance from a grocery store 2) the amount of gas in your
tank (or rubber on the soles of your shoes, depending on how
you get to the store) 3) noise level in the kitchen (as you
unload the groceries) etc. etc.

Best

Rick

···

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

[From Bruce Nevin (991019.1515 EDT)]

Rick Marken (991019.1100)

Bruce Nevin (991019.1042 EDT)--

A loop controlling a tactile pressure intensity perception
could possibly have many side effects if a finger is pushing
a button or a key on a keyboard, but this is exceptional.

All control loops produce many (probably an infinite number
of) side effects. Your example of a system controlling the amount
of groceries in the kitchen will produce a zillion side effects;
these side effects are effects (on _any_ variables) produced by
the actions the system takes to maintain the amount groceries
at the desired level. So, as this system acts to control the
amount of groceries, it will affect variables like 1) your
distance from a grocery store 2) the amount of gas in your
tank (or rubber on the soles of your shoes, depending on how
you get to the store) 3) noise level in the kitchen (as you
unload the groceries) etc. etc.

Thanks for reiterating half of the point.

The other half is that there are fewer side effects lower in the hierarchy.

Suppose as part of controlling "kitchen has enough groceries stored" you
unlatch a door, and among the means for doing that you control a
pressure-intensity perception (ah, it's unlatched now, I can stop turning
the knob and start pulling). All the side effects of controlling that
intensity perception are a small fraction of the "zillion" side effects of
controlling "kitchen has enough groceries stored". None of the side effects
of e.g. lifting the grocery bag after opening the door or saying "Oh, OK,
there's room in here for it" are included among the side effects of
controlling that intensity perception.

On another occasion, controlling the perceptual signal from that same
intensity receptor (or more likely a group of them) could have the
consequence of setting off a bomb that destroys a building and kills many
people, or of sending an email message that some fruitcake takes as an
excuse to hunt me down and throw a cream pie in my face, but such scenarios
involve other control systems so I think we can set them aside, and that
was the point of the 3 lines that you quoted.

  Bruce Nevin

  Bruce Nevin

···

At 10:57 AM 10/19/1999 -0700, Richard Marken wrote:

[From Rick Marken (991019.1255)]

Bruce Nevin (991019.1515 EDT)--

Thanks for reiterating half of the point.

The other half is that there are fewer side effects lower
in the hierarchy.

I don't believe this. My intuition is that there is the
same (infinite) number of side effects regardless of what
kind of variable (level in the hierarchy) is being controlled.
Only a quantitative enumeration of the side effects associated
with control of a "higher level" and a "lower level" perception
would convince me that there is any difference in the number
of side effects associated with each. Your claim that there
are "fewer side effects lower in the hierarchy" sounds like
the kind of thing I used to do with my brother; I'd make up
authoritative answers to all his questions; I think he still
believes that a watched pot never boils and that there are
fewer side effects lower in the hierarchy;-)

controlling the perceptual signal from that same intensity
receptor (or more likely a group of them) could have the
consequence of setting off a bomb that destroys a building
and kills many people, or of sending an email message that
some fruitcake takes as an excuse to hunt me down and throw
a cream pie in my face, but such scenarios involve other
control systems so I think we can set them aside, and that
was the point of the 3 lines that you quoted.

I don't think you quite understand the concept of _side effect_.
A side effect is not just an unintended event (like a bomb
blast) that is caused by a control action. It is variation in
_any_ measurable (perceptible) aspect of the world that is
influenced by the action. That's why I say that any control
action will have a zillion (more precisely, an infinite) number
of different side effects. Here is a partial list of the
side effects (variables influenced by movement of your finger)
that occur when you control the pressure at your fingertip:

1) distance between your fingertip and every object in the
room

2) air pressure at your fingertip

3) noise level in the room

4) rate of movement of button

5) blood pressure in fingertip

6) temperature at surface of button being pressed

7) etc.

Does this help?

Best

Rick

···

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

[From ]

Rick Marken (991019.1255)

A side effect [...] is variation in
_any_ measurable (perceptible) aspect of the world that is
influenced by the action. That's why I say that any control
action will have a zillion (more precisely, an infinite) number
of different side effects. Here is a partial list of the
side effects (variables influenced by movement of your finger)
that occur when you control the pressure at your fingertip:

[...]

Does this help?

What I said:

It follows that at lower levels there can be more side effects,
at higher levels, fewer.

I can amend that:

It follows that the side effects of a control loop are a subset of the side
effects of any higher-level control loop that is using it as a means of
control.

Is that better?

I won't argue infinite sets with you. You quickly reach verbal absurdities.
When infinite set A (e.g. the set of even numbers) is a subset of infinite
set B (e.g. the set of natural numbers, even and odd) is it smaller? If
not, why not? For every member of the first set there are two of the
second. Who cares if you can never find the end of either set. For almost
all purposes, infinities are neither relevant nor useful. At any relevant
or useful point in your counting there are twice as many of one as there
are of the other, and there always will be, throughout an infinitely long
counting process. (Mathematicians since Cantor speak of infinities of
different cardinality to distinguish infinities of different sizes, as I
recall, but my memory is fuzzy.)

I am assuming that every control loop is closed through the physical
environment. (The alternative I am rejecting is to say that the
"environment" of a configuration detector, say, is a universe of sensation
perceptions presented to it from the next level down, not an easy
hypothesis to support.) When I control a perception above the level of
intensities, it is by means of other control loops at lower levels of the
hierarchy. On the above assumption, then, the lower control loops that are
being used as means of control by the higher one are part of the higher
control loop. It follows that the side effects of any lower-level control
loop are going to be a subset of the side effects of the entire
higher-level control loop in which it participates. (An alternative would
be to say that side effects at a given level must be perceptions at that
same level. This would be to say that bumping you with the grocery cart
would not be a side effect of supplying the kitchen with groceries, nor
even of directing the cart into aisle 7, it would be a side effect of
pushing the cart. I'll leave that line of argument to you if you want it.)

Yes, you can say that each loop's side effects are an infinite set. But the
one set of side effects is a subset of the other. And certainly the set of
relevant, or disturbing, or noticeable side effects is smaller at the lower
level and (accumulating many such smaller sets) larger at the higher level.

  Bruce Nevin

···

At 12:57 PM 10/19/1999 -0700, Richard Marken wrote:

[From Rick Marken (991019.1520)]

Bruce Nevin says:

It follows that the side effects of a control loop are a
subset of the side effects of any higher-level control loop
that is using it as a means of control.

Is that better?

Nah. Still don't buy it. But it doesn't matter anyway. As
long as we agree that an infinite number of side effects
are produced when controlling _any_ variable, then it really
doesn't matter to me if you want to imagine that the infinity
of side effects for some these variables is bigger than the
infinity for others.

Best

Rick

···

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

[Martin Taylor 991020 10:20]

[From ]

I can amend that:

It follows that the side effects of a control loop are a subset of the side
effects of any higher-level control loop that is using it as a means of
control.

Is that better?

No, it doesn't follow from the model I described and which you seem not
to like.

In my model, the intended effect of a high-level control loop is
achieved by executing one or more low-level control loops in such a
way that the side-effects of the low-level actions directly influence
the high-level perception. Those useful side-effects are _not_ in a
subset of the side-effects of the high-level loop. Here's a HIGHLY
simplified diagram that omits boxes for Perceptual Input functions
and output functions.

                       >
                       >
            ---->------O--->------+-----------------------------
           > V V
     intermediate | |
        control ----->----O---->----- other
         loops | | low-level
           ^ ^ |actions control
           > > > loops
==========^===========^=====================V================= V ======
           > > > >
           > +-----------<---------+---->other--X +----X
           ^ |disturbances | side |
           > ^ V effects |
           > > >
           +--------<---------------<-------- |
           > "useful" side-effects |
           +------<-------------<---------------<---------------
           >
      other>disturbances

Some side effects of the actions of the lower-level control loops are
also side effects of the action of the high-level loop; some are not.
The side-effects marked "X" are subsets of the set of side-effects of
the actions of the high-level loop. The ones marked "useful" are not.

Note that in the diagram, the perceptual signal of the expanded lower
loop does _not_ contribute to the perceptual signal of the higher-level
loop. This is what makes the influence of its actions on the high-level
perception into a side-effect. Usually the drawings of HPCT show the
perceptual signals of low-level control units going back to the high-level
units that provided their reference signals. But this need not be the
case, and cannot be the case at the lowest levels, where muscle-control
loops are clearly not sending their perceptual signals back, for example,
to the visual system or the olfactory system.

What I am saying is twofold: (1) This kind of side-effect connection
is prevalent in a randomly organized set of control loops that work
through a common external environment, and (2) the simplest kind of
reorganization theory produces the kind of structures shown in the
diagram. According to the theory, "helpful" side-effect connections are
retained, while "disturbing" or "hurtful" connections are lost.

Given that such connections among control loops almost inevitably occur,
it is a reasonable suggestion that they are responsible for some of the
cases in which it looks as if low-level "error" induces high-level
control loops to act. (The effective error is in the high-level control
system that set the reference for the low-level one experiencing
error, but I couldn't show that in this diagram, given the space
available.)

I hope this clarifies more than it obfuscates.

Martin