Perceptual Control; Association; ThreeCV1 update

[From Bruce Abbott (950114.1930 EST)]

Greetings from TROPICAL Fort Wayne! This is totally amazing: walking
around in the Midwest in mid-January without need of a coat, and the grass
still green. By gosh, I think I'm going to LIKE global warming . . .

Bill Powers (950114.0750 MST) --

On Habits:

The main thing to remember about complex habits is that they are
sequences of controlled perceptions, not sequences of actions. They
therefore result from turning on a reference signal that specifies a
familiar sequence of perceptions. What repeats at that level of
perception usually requires reference signals and perceptions at lower
levels that do NOT repeat, because the environment is never the same
twice. When you found yourself "making a pot of coffee," were the
filters exactly where they were before, did you peel one off the stack
(the same filter?) using exactly the same muscle tensions, finger
positions, arm positions, and body orientation that you used the last
time? If somehow you had been able to record the signals entering your
muscles the last time you did this, and played them back into the
muscles with complete accuracy, do you think the result would have been
a new pot of coffee? Did you, perchance, have to pour out the old coffee
and rinse the pot before starting the new pot?

I tried to make it clear that I was speaking of controlled perceptions. In
my post I wrote:

Which brings me to a question: what role should association play in PCT?
Obviously the traditional notion of skilled execution as a stimulus-response
chain won't wash, but it seems to me that realization of one goal in a


of goals that must be accomplished could be the "switch" that passes

control to

the next control system in the sequence.

"Realization of one goal" equals bringing the perception under control by
that system to its reference level. You suggest that the program-level
control system switches on the next lower-level system in the "program" by
raising its reference level. What I am asking is whether you see any use
for associative principles to account for the "construction" of such
sequences. I'll try to develop this idea further by elaborating my example.

When I left my office to get that coffee cup, I had a high-level goal
consciously in mind (if getting rewarmed coffee can be called high level),
one that required my to accomplish a series of subgoals: getting out of my
chair, turning toward the door, moving to the next room, and so on. There
may in fact have been several ways I could have accomplished each of these
goals (I could have skipped into the next room rather than walking, for
example). As you so nicely pointed out, I am not really planning a set of
movements but rather a series of perceptions. Now, how is it that the
program control level "chooses" one set of subgoals over another, perhaps
equally feasible set? At times we may be consciously aware of debating
alternatives and choosing among them; at other times we just "mindlessly"
let the lower-level control systems pick. Some things have been done so
often they have become habits and we can go through the whole series without
much conscious thought. Walter Schneider likes to distinguish those things
requiring conscious intervention from those which have become habitual as
"controlled" versus "automatic" processes. The choice of the term
"controlled" is unfortunate; what he really means is that some processes
involve higher-level ("conscious") control systems than do others.

This higher-level control system appears able to create a novel sequence of
subgoals denovo in order to being about some higher goal. It would seem that
by consciously repeating that sequence, we gradually lay down a pattern, not
of behaviors, but of goals to be accomplished in some fixed sequence, which
in effect "trains" the correct reference output sequence into the
program-level control system. An interesting question for PCT, it seems to
me, is how this sequence gets organized. When we learn a habit, are we in
some way laying down a set of associative connections such that the
achievement of one perceptual goal triggers the raising of the next
reference level?

I realize that in asking this question I am probably getting far ahead of
the data; I'm mainly interested in whether you see the concept of
association as at all useful. There's plenty of evidence that memory is
associative in the purely descriptive meaning of the term, and it may well
be that, like S-R, the appearances are deceiving. Still, there must be SOME
mechanism at work to provide the appearances, and I'd be interested in your
thoughts on the matter.


I've revised THREECV1 so as to write the collected data to a file, and have
moved the PCT analysis stuff into a separate program that reads the data
file. This way, I can give a copy of the data to any of my colleagues who
wish to try their hand at a conventional analysis.

The revised THREECV1 examines the current default directory for a data file
called PCTDATA.000; finding none, it uses this as the filename. If such a
file exists, it increments the extension to 001 and repeats the sequence;
this process continues until a unique filename is found. This scheme allows
the program to create 1000 uniquely-named files before it fails, which
should be more than adequate. I'm planning to put THREECV1 on our Neff_Labs
server and have my statistics classes generate some data, which will be
saved to the server's hard disk. (I already have a reaction-time program
installed there that saves data in this way.)

The data files are ascii so that they can be easily imported into various
statistical packages or otherwise manipulated. A data file contains seven
values per line (three disturbances, three cursor positions, and the mouse
position) and 3600 lines, so these are fairly large files. I discovered
that Minitab as configured for my PC ran out of room after reading 2370 rows
and refused to read the rest. Even then, I had to delete some variables
while analyzing others because there was insufficient workspace for the
calculations unless I did so.

One way to shrink the files would be to eliminate saving the three cursor
positions, as these can be reconstructed from the mouse position and
disturbances. Do you think this would be a problem? Would it be better to
omit the disturbances and keep the cursor positions? What I'm concerned
about is not giving my colleagues a data set that suggests a particular way
of approaching the analysis.

In taking a look at my own data via Minitab (the portion that loaded), my
performance seems fairly typical. The correlations between disturbance and
mouse position were -.033, -.040, and -.992. Guess which cursor I was
controlling? The correlations between cursor position and mouse position
were .727, -.358, and .133. Correlating disturbance 3 at lag 10 with mouse
position raised the correlation from -.992 to -.996, suggesting that my
control lag was about 10/60 or 1/6 second, if I understand the time units

Tom Bourbon [950112.1607] mentioned to Martin Taylor the possibility of
using a routine to mark a portion of his data from a run for separate
analysis; this is easy to do with the THREECV1 ascii files: just cut and paste.

I still have a bit of work to do to "clean up" the revised programs; I'll
post them when they seem to be in good order.

Are we having fun yet?