[From Bill Powers (970919.162MDT)]
Rupert Young (970919.18.30 BST)--
Oh god, just when I thought I've understood PCT I start thinking about it
and realise I'm still thinking in S-R and have no idea what's going on.
The beginning of wisdom.
One thing is how can you recognise something when you are controlling for
another variable?
This is a "massively parallel" system. Hypothetically anyway, all the
perceptual functions at all the levels are producing perceptual signals all
of the time, with the largest perceptual signals being those from systems
whose inputs come closest to the vector defined by their input functions. I
suspect that Martin Taylor could explain this more clearly.
Because of the nature of neural signals, they can't go negative. This means
that if the reference signal is excitatory and the perceptual signal
inhibitory at the comparator, a zero reference signal will result in zero
error no matter what the perceptual signal is. So setting the reference
signal to zero turns the control system off: it will not act. However, its
perceptual signal still goes to the next level up, contributing to the
total input to that next level.
If you're "looking for" something, the reference signal for that perception
is set to a nonzero level. Error signals exist, and the output of the
system acts on the world to bring that perception to the desired value. At
the lowest levels this may amount to nothing more than bringing an image to
the center of vision. Other images elsewhere on the retina may still
produce perceptual signals in other perceptual functions, but they simply
pass the information on upward because they are not part of active control
systems.
I suspect that object recognition needs images to be foveated. But even a
foveated image presents arrays of sensations that can give rise to more
than one configuration perception at the same time. I'm looking at my
glasses case, but I can also see a floppy disk in the same central field.
But if I'm looking _for_ my glasses case, it's that perception that's
actively controlled (by picking it up). The floppy disk is just there.
At the sensation level there aren't many actions that can control them. You
can move your eyes, cock your head so your ears hear better, press foods
with your tongue to bring up their taste, and (mostly) tense your muscles
to create sensations of effort. Sensation and intensity control are
involved most directly with the kinesthetic control systems. Most sensation
signals and intensity signals are not part of low-level control systems,
although most of then _can_ be controlled in special situations. There has
to be some kind of external connection that allows muscle efforts to alter
them.
The hardest part of understanding low-level control (that is, as well as
anyone understands it) is to avoid using higher-level interpretations.
I can see that the input to the sensation level is a collection of intensity
control system perceptions and the input to the configuration level is a
collection of sensation control system perceptions, but what about the
output? 'm still thinking in feed-forward mode. Does the output from the
higher ontrol system go to the same lower systems from which it got its
input ?
The output from a higher control systems sets a reference level for lower
control systems, telling them what to perceive. But the input that the
higher system gets comes, in general, both from controlled and uncontrolled
lower perceptions. The uncontrolled perceptions can vary as the environment
changes. This disturbs the higher system, and it adjusts the reference
signals for the lower system, altering those lower perceptual signals that
_are_ under control to compensate for the changes in the uncontrolled inputs.
If you want to try modeling this, build a two-level system as follows:
At level 2, the control system receives two level-one perceptual signals
p1a and p1b, and the second-level perceptual signal represents their sum:
p2 = p1a + p1b. The error signal from the second-level system sets the
reference signal for a single control system at level 1 controlling p1a;
p1b comes from the environment and is uncontrolled. Let p1a come from
environmental variable v1, and p1b come from v2. The output of the
first-level control system affects v2, but not v1. You can set v1
arbitrarily, and vary it.
What you will find is that the perceptual signal p2 remains near whatever
reference level the higher system is given. The higher system adjusts the
reference signal for the lower system so that p1b, when added to p1a,
provides the correct amount of p2. In fact, when you vary v1 to cause p1a
to change, p1b changes the opposite way: for the second-level system,
changes in p1a caused by changes in v1 are just disturbances.
In a tracking situation we have exactly this case. The target is perceived
against a background, and so is the cursor. The participant has a way of
controlling the perceived cursor position, but the perceived target
position is uncontrolled -- it varies independently. The task is to control
a _relationship_ between these two lower-level perceptions, by varying the
perception that is directly controllable to compensate for changes in the
other perception. That's exactly what happens.
Now imagine that the second-level input function receives many perceptual
signals from level 1; some are controlled and the rest are uncontrolled.
The controlled perceptions (adjusted by varying the reference signals for
all the first-level control systems) are adjusted to compensate for changes
in the uncontrolled ones, so the second-level perception remains
essentially undisturbed. The set of all uncontrolled perceptions that
contribute to the higher-level perception is the effective disturbance of
the higher-level perception; the controlled inputs are adjusted to maintain
the higher-level perception at the reference level.
This same general relationship hold between any pair of adjacent levels.
Some of the signals entering the higher perceptual function are controlled;
the rest are not. So we can define disturbances that act at any level.
I hope this helps.
Best,
Bill P.