[From Bruce Abbott (991101.1800 EST)]
Rick Marken (991101.2200) --
A control model does this by controlling a _perception_ of a
sequence of tasks; this would not necessarily involve producing
a sequence of behaviors (what you call "control tasks").
Bruce Abbott (991101.1940 EST)
Great, then I'll ask you the same question I asked Bruce Gregory:
How does a sequence-control system "know" when to switch from
controlling one perception (by means of one control system) to
controlling an entirely different perception (by means of a
different control system)?
A sequence control system doesn't have to "know" when to
switch from controlling one perception to controlling another
because it doesn't switch from controlling one perception to
controlling another: it is _always_ controlling the same
perception; a sequence perception. If you can understand this,
you will understand control of perceptions, including perceptions
like sequences and programs, that a defined over time.
A sequence control system (like all perceptual control
systems) controls a unidimensional signal that is the
perceptual signal. In a sequence control system the
intensity of this signal indicates the degree to which
the sequence is present in lower level perceptions.
Examples of perceptual functions that can produce a signal
whose intensity represents the degree to which a sequence
is occurring can be found in Uncle Bill's B:CP p. 144.
I think you have in mind something like your demo in which particular
sequences of changes in, say, the shape of an object take place, and these
sequences change from time to time. By responding appropriately, you are
able to change the sequence currently being shown back to a particular one
specified as the reference sequence. According to theory, you have
constructed (somehow) a sequence perceiver that outputs a signal whose
intensity is proportional to the "degree to which the sequence is present."
If I understand Bill's sequence-perceiving mechanism correctly, it is
basically a set of flip-flops that must be set in sequence by the arrivals,
in proper sequence, of input signals representing the presence of the
required state of each separate perception in the sequence. As each
flip-flop in the series turns on, this event indicates that the required
goal-state associated with that flip flop, plus the required goal-states
associated with all previous flip-flops in the series, have occurred, and
occurred in the required order. Evidently there would be an output signal
that would remain at a given intensity so long as each perception in the
sequence occurred according to the sequence specified. If the specified
perceptual inputs did not occur in proper order, the output signal's
intensity would change, creating a mismatch between the intensity of the
sequence perceptual signal and its reference and initiating some sort of
corrective action (e.g., pressing a button to change the sequence,
observing the sequence again, and repeating until the specified sequence
appears -- the e. coli strategy).
Such a mechanism can be made to work, but I rather doubt that it represents
how I actually go about the task. When I pay attention to what I'm doing
while performing such a task, it seems to me that I perceive, say the circle
(first element in the sequence) and then look for the square (second element
in the sequence), and so on. I don't get the impression that I'm perceiving
a signal whose intensity varies with the order in which these occur. At
each step I perceive a match between what I expect and what I see, or I
perceive a mismatch. What I expect to appear next seems to be the result of
a process that has linked a set of reference specifications together in
memory, so that when I see one, this perception produces the recall of the next.
Now let's look at a different situation. You are attempting to _produce_ a
particular sequence of perceptions, not just _observe_ a sequence and decide
whether it matches a specified sequence. It is no longer a matter of just
watching perceptions come in, one after the other, and then taking action if
the observed sequence fails to maintain a given level of output from a
sequence-perceiver. Now, in the current model, you need more than just a
sequence perceiver. You need a whole series of control systems each
organized to produce a particular perception that is required as part of the
sequence, and these must be organized so that they come into play in the
required order. Even the nature of the perception to be controlled may
differ across the sequence. Appropriate actions must be initiated as
required to bring each of these perceputal inputs to its reference state in
sequence order, first one, then the next, then the next after that. As each
act is completed, the sequence-perceiver must not only ratchet to the next
input in the series, it must deactivate the current control system and
activate the next.
I hope you can see now that just because you can see
a rat carrying out a sequence of "control tasks" doesn't
mean that the rat is controlling a sequence perception.
Rick, are you even _aware_ that this whole discussion began with my
attempting to show how a rat could produce a specific sequence of
perceptions without controlling a sequence perception? Your comment
indicates that you are not.
And a rat who actually _is_ controlling a sequence
perception may be doing so without actually carrying
out a sequence of "control tasks" (for example, if Chris
Cherpas' rats were controlling a sequence of colors, they
were doing it without carrying out the same sequence of
tasks (pecks) each time the perceptual sequence was produced.
The study Chris described used pigeons, not rats. (Rats cannot perceive
colors.) His description of the procedure was somewhat vague -- it is not
clear to me what the requirement was for "pecking on a color" once that
color was made to appear, or what would induce the pigeon to return to the
changeover key to select another color. What if the task involved selecting
blue, then green, then red, before reinforcement would be delivered? The
pigeon would have to set a reference for blue at the start, then green,
given that blue was present, then red, given that green was present. Each
color provides the retrieval cue for remembering the next color to produce,
and it is then a mere matter of matching the color showing on the key to the
Rick, you suggest that under my proposal the "sequence of tasks" would have
to be the same right down to the individual keypecks, and that the pigeons
in this experiment were violating this requirement. But my proposal imposes
no such requirement. The actions produced to bring about each reference
state in the sequence can vary as necessary, as they are the actions of
control systems. Completion of the first control task produces a cue which
determines which new control task should now be selected (in this case,
which color should now be produced on the color key).