NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/bit/listserv/csgl/709

Text File | 1992-08-16 | 8.5 KB | 152 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!gatech!paladin.american.edu!auvm!VAXF.COLORADO.EDU!POWERS_W
X-Envelope-to: [e-mail]
X-VMS-To: @CSG
MIME-version: 1.0
Content-transfer-encoding: 7BIT
Message-ID: <[e-mail]>
Newsgroups: bit.listserv.csg-l
Date:         Sun, 16 Aug 1992 16:48:38 -0600
Sender:       "Control Systems Group Network (CSGnet)" <[e-mail]>
From:         "William T. Powers" <[e-mail]>
Subject:      Why control
X-To:         [e-mail]
Lines: 136

[From Bill Powers (920816.1530)]

Avery Andrews (920816) --

You make a nice case for PCT, but it could be carried further. This:

>What I'd say to this is that PCT is not cognitivist, because the
>*phenomenon* of control is only found when an organism (or other >`Agent')
is put in an appropriate environment (e.g. if the road is >covered with an
oil-slick, one ceases to observe control of the >direction in which the car
is going).

... could be expanded by contrast with the environments that Pengi and
Sonja are given. The whole digital approach to imitating behavior (Rick's
meaning of Imitating) is based on the expectation that there is a reliable
causal chain all the way from the top command level to the final
environmental result. Peripheral devices used in digital control systems
are designed like the computer itself -- an output pulse causes a stepper
motor to execute exactly 7.5 degrees (or whatever) of rotation, and the
gear or screw on the shaft causes the load to advance or retract by an
exactly known and repeatable amount. On the input end, you type a key on a
keyboard or receive a modem code, and never get anything partway between A
and B. Any error whatsoever is likely to be fatal. Sonja uses a joystick
that always produces a motion in one of eight directions -- exactly. When
Pengi kicks an ice-cube, the ice-cube always goes along the lines of a
grid. And so on. This is an environment in which the lowest levels of
action are absolutely precise, predictable, and undisturbed. That's the
ONLY reason that these programs can work.

If you substituted a normal continuously-variable environment with
continuously-variable disturbances (even moderate and slow ones), and if
the means of action were also made continuously variable, there would be no
way to reduce the production of action to if-then rules; do it or don't do
it. I think this would scuttle most AI and AL programs.

When I talk about disturbances, I don't mean big dramatic things like an
oil-spill on the road. I mean just tiny little things: a gust of air, a
pebble that rolls a bit underfoot, a muscle that gets a little tired, a
slightly awkward stance when you aim, a little overshoot when you whip
around a corner, a bit of play in the steering gear. Most action is a
succession of events that begin where the previous event left off; we move
not in terms of position but in terms of accelerations and velocities which
integrate to positions. Every move we make is slightly disturbed, a little
different from what was intended, a little off the mark. If it weren't for
continuous control operating all the time at many levels of organization,
these errors would be cumulative. We would blunder around bumping into
doorframes, stumbling over our own feet, knocking glasses of water over
instead of picking them up, driving off the road and across the fields, and
crushing or dropping eggs when we tried to pick them up. The only thing
that allows behavior to work at all is control, even in a quite normal
environment.

And control is not a matter of commanding outputs, but of requesting
inputs. The outputs vary according to whatever disturbances are happening,
regardless of the goal. Only the closed-loop control organization can do
this.

>There is a terminological difficulty here in that in normal usage,
>`control' does *not* imply any capacity to cope with unpredictable
>disturbances as they arise (I have retained a copy of one of Randy >Beer's
postings that documents this point extensively, which I could >send to
anyone who'se interested in looking at it).

This terminological difficulty arose because people (including Randy Beer)
don't understand how control works. In most usages of this term, the user
assumes that setting up the "controlling" conditions is enough to assure
the "controlled" result. The role of the controlling entity in the
background is overlooked. When we speak of the "tone control" on a radio,
we mean that this knob can set the tone to any state between full bass and
full treble -- provided that there is a person using it for that purpose,
relative to a desired tone. Left to itself, the tone control knob can't
control anything. If the speaker is muffled by a towel, the tone control
won't turn up the treble. When it's said that temperature controls the rate
of a chemical reaction, the same thing applies; the rate of the reaction is
controlled by someone's varying the temperature as required to produce a
given reaction rate. If the reaction rate varies because of changes in the
chemistry, the temperature, left to itself, won't do anything about it.

All such uses of "control" carry the unspoken addition: "all else being
equal." But this is a false interpretation of control; control does not
require all else to be equal. When control works -- when you make the rate
of the reaction constant by adjusting the temperature -- the rate of the
reaction is very reliably controlled, even if you vary the temperature in
order to keep the reaction rate constant and even if external factors that
affect reaction rate require doing this.

Because people don't understand control, they think that simply adjusting
the MEANS of control is sufficient to produce a reliable result. They
believe this in part because when they see the same result being repeated
or maintained, they assume that "all else" MUST have been equal, when in
fact it was not. They see control taking place, and don't realize that
sensing the result is absolutely essential to success. They don't realize
that the adjustments taking place are necessary because of ongoing
disturbances. They see someone set the knob to 20 and think that in the
future, all they have to do is set the knob to 20 again to get the same
result again. Even when they do the controlling themselves, they are
unaware of the role their own perceptions are playing.

I've told this before, but it's worth a repeat in this context. A
psychologist once insisted to me that to drive a car around a bend in the
road, all he did was turn the wheel by the right amount, and the car just
went around the curve. He didn't see any reason to watch the road during
the turn. I don't know if he's still alive.

When people talk about controlling something by some means, they mean that
the something is reliably produced. They don't mean that the controlling
circumstances are established and the result varies all over the place.
They just aren't aware that in most circumstances, if there were no
feedback control involved, setting up the controlling circumstances and
leaving them set WOULD allow the result to vary all over the place. Even
when setting a control will produce a reliable result for some period of
time, they aren't aware of that little feedback loop that works while the
control is being set. They aren't watching the controlling device, but the
result -- that tells them when the adjustment is "right". They feel the
bathwater while turning the hot and cold water knobs. They watch the flame
while they adjust the burner on the stove. They feel the resistance build
up as they pull the hand brake to "on." But when they describe these
control processes, they describe the means: the turning of the faucet, the
turning of the knob on the stove, the pulling on the hand brake.

I claim that in just about every case where "control" is used in a way
different from that in PCT, the person using the term really expects true
control to happen. The outcome is expected to come to a particular state
and stay there. But the action or situation that is spoken of as
controlling is hardly ever sufficient to produce that result. What is
usually meant is "influence" or "affect" or (all else being equal)
"determine" . That is not sufficient to produce the expected result of
constancy or stability, save in a highly artificial environment constructed
to exclude all disturbances and allow only discrete settings of the
"controlling" element. And even then, someone must find the adjustment that
produces the desired outcome, not by watching the adjustment but by
watching the outcome.

Best

Bill P.