Manual control

[From Rick Marken (930920.2100)]

Dag Forssell (930920 1840)

I am not familiar with "manual control." Could you enlighten me?
Whom are you quoting? What is the relevance to PCT? Are you
talking about control engineering?

"Manual control" refers to the work of a group of psychologists who
have been applying control theory since the at least mid 1950s to
continuous control tasks, like tracking. They get the control theory
part fine (as I said) but they make some mistakes when they apply
it to living systems (not all do, as Greg Williams showed me at the
meeting -- but a hefty majority, from my experience). Take a pursuit
tracking task, for example, where the subject is asked to keep a
cursor (c) aligned with a moving target (t). In manual control,
this is modelled as follows:

                         Person

ยทยทยท

___________________________________
          > >
t--> e -->| processing --> more processing -->| -->o
     ^ ____________________________________ |
     > >
     c<---------------------------------------------

The difference between target and cursor is computed OUTSIDE THE PERSON;
the computed error (t-c) is the thus the INPUT variable. This eliminates
the need to think about reference states for t-c. The input is processed in
all kinds of complex ways resulting, finally, in the output (handle
movement) that influences the cursor (and, hence the input, t-c). The
manual controllers study the dynamics (frequency response, time lags,
etc) in the relationship between e (or t) and o. Of course, e is really
a controlled variable and t is a disturbance to this variable. So the
relationship between t and o is the behavioral illusion -- it seems
like the relationship between t and o would reflect characteristics
of the processing in the subject; in fact, it reflects just the
characteristics of the feedback function relating o to e via c.

If they take e to be the input then they are taking the controlled
variable as the cause of output, o. Now they're looking at something
a bit more like my version of the behavioral illusion. Fairly
strong relationships between e and o are found in these studies
because the manual controllers see a LOT of variance in e; this
is because they use pretty high frequency, high amplitude dis-
turbances, making e pretty tough to control.

The manual controllers are studying what many PCTers are studying--
control -- but the manual controllers are looking at control from
a different perspective. What fascinates the manual controllers is
how inputs (disturbances, "errors") seem to control outputs;
what fascinates PCTers (as you know) is how people vary their
outputs as necessary in order to control inputs RELATIVE
TO THEIR OWN SPECIFICATIONS FOR THESE INPUTS.

Best

Rick

[Avery Andrews 930921.1505]
(Rick Marken (930920.2100))

Perhaps the reason that the `manual controllers' are uninterested in r is that
in their applications, it is obvious what r is, so there is no need
to figure out how to identify it. Similarly with control systems
engineering, as discussed by Hans Blom.

Avery.Andrews@anu.edu.au

[From Rick Marken (930921.0730)]

Avery Andrews (930921.1505) --

Perhaps the reason that the `manual controllers' are uninterested in r is

that

in their applications, it is obvious what r is, so there is no need
to figure out how to identify it. Similarly with control systems
engineering, as discussed by Hans Blom.

Right on.

Not only is r obvious in these tasks; what is also pretty obvious is
the environmental correlate of the perceptual variable being controlled.
So the 'manual controllers' and control engineers know both p and the
reference setting for p, which is r. In a tracking task, for example, p
is obviously the difference between target and cursor positions; r is
obviously zero (obvious because the difference between target and cursor
is kept NEARLY at zero despite disturbances to this variable). In control
engineering, p is the measured value of the "to be controlled" variable --
temperature, density, level of liquid, etc; r is the user-specified level
of this variable.

Of course, determining the environmental correlate of p (as well as the
prevailing reference specification for this variable, r) is what "the
test" is all about. Thus, "the test" is aimed at discovering things that
'manual controllers' and control systems engineers couldn't possibly see
a need for discovering -- except under very special circumstances. If we
could get 'manual controllers', for example, to watch while a subject
controls the difference between cursor and target position while
continuously changing the reference for that distance, we might be able
to get them to see that r is actually determined inside the subject,
not in the environment. More importantly, if we could get them to
watch while the subject switches from controlling the shape of a
quadralateral to controlling its size (area) then we might be able to
get them to see that different aspects of the same objective situation
can be controlled; then we might be able to convince them that what
is controlled is what is perceived; and, finally, we might be able to
make a case for doing "the test" by noting that it is generally impossible
to tell (just by looking) what aspects of an objective situation are
under control.

Best

Rick