[From Bill Powers (950114.0750 MST)]
Avery Andrews (950113) --
The hand control of speech project sounds very appealing. The offer for
the project with Bruce Nevin still holds, when you come up with a
specific proposal.
RE: degrees of freedom:
The degrees of freedom problem is also a crock - it's perfectly
easy to use 2nd order feedback systems (like in Rick's
spreadsheetmodels, and my `nu14' program, to manage `excess'
degrees of freedom ('nobody ever complained about having too many
degrees of freedom' says a famous Russian motor person now at
McGill whose name I can't remember right now, who I think is much
more on track that Bizzi and his mates).
The actual problem is that these `excess' df's are actually
resources which you can apply in various ways in different
situations
I think you've got the right view on degrees of freedom. Somehow the
notion has arisen that the environment (and perception) contains so many
degrees of freedom that they overwhelm the available output degrees of
freedom, as if the organism's duty is to control the environment as well
as possible in all possible ways. But the organism learns to perceive
everything it can control as an independent entity. The "excess" dfs
amount to don't-care conditions.
···
-----------------------------------------------------------------------
Gary Cziko (950113.1733 GMT)--
Some Bill Powers's posts are there, but I can't find all of them and
don't know why.
Ne neither. We did have a problem like this some time in the past, and I
believe it was fixed at your end. So I don't remember the solution.
-----------------------------------------------------------------------
Bruce Abbott (950113.1245 EST)--
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 don't think that metaphors like "paths through familiar degrees of
freedom" are very helpful.
It's not surprising that you carried out a sequence that you had no
conscious intention of carrying out. Very few intentions are conscious.
You witnessed a small part of the operation of a system that is much
larger than your field of consciousness, and which you can consciously
affect only in scattered places.
When you use your thumb to create a space on the screen, are you
conscious of the intention that your thumb will move the perception of a
space bar down until a certain degree of resistance is felt? While
you're composing a comment on someone's post, are you conscious of the
intention to be part of an internet discussion? Your attention can be
called to such intentions, but they exist and are effective whether or
not you are attending to them.
I think one property of sequence control is that when we perform the
first element of the sequence, the whole sequence tends to play itself
out. So if two sequences begin with the same element, it's possible that
a sequence will start that wasn't the one you consciously intended --
although obviously there was an intention to produce the other sequence,
too. Perhaps if the reference level for a given sequence is zero,
occurrance of the first element constitutes an error, and the only way
to correct the error is to go all the way through the sequence to the
state where the sequence is once more not occurring. Or perhaps some
system at the level where you are currently conscious has a goal
different from the one of which you are aware: maybe there's another
system that prefers fresh coffee, however uneconomical it seems to a
different system to throw away the old coffee.
Habits are simply recognizeable control processes. They're usually
defined in terms of consequences of actions, not the actions themselves
that create the consequences. When you see a pattern of behavior
occurring at one level of organization, you can be pretty sure it isn't
repeating at any lower level.
It does seem that following well-worn pathways to a goal would serve to
reduce the need to deal with "excess" degrees of freedom, as Bruce
[Nevin] suggests. Research project, anyone?
There are no "well-worn pathways" of action leading to a goal. That's
still another metaphor, this time one that grows out of assuming a
constant environment. The fact that a consequence repeats is no
indication that the actions which brought it about repeated. However, it
is true that there are causal relations in the environment of which we
take advantage: if we find that pushing on a door opens it, we are
likely to push on that door each time we intend that it be open. The
"well worn pathway" to the goal of starting a car's engine entails
perceiving the ignition key in the lock and feeling a certain twisting
effort, and hearing a certain sound that indicates that the engine has
started running. However, all this can be done with either hand, sitting
in either seat, with the body in any orientation, and so forth. The
well-worn pathway is not in the nervous system, but in the causal
relations in the environment capable of creating the perception we want
to experience. In most cases there are alternative causal relationships
we might use, and do use when disturbances require, but learning stops
when the perception is controlled, not necessarily when the best
possible method of controlling it is found.
----------------------------------------------------------------------
Martin Taylor (950113.1615) --
The classic "conflict" comes about because there are two independent
perceptual signals, but at one point in the two loops, the pathways
share one single degree of freedom, the common CEV.
This is where Avery Andrews' comment comes into play. The objective is
not to use particular lower-level degrees of freedom, but to control a
given perception at a higher level. If a df conflict occurs, the obvious
thing to do is change the lower systems being used to eliminate the
conflict while still permitting control of the higher perceptions.
Thinking of the CEV as existing in the environment suggests that we're
stuck with a particular way of defining that CEV. But when we recognize
that it is a higher-level perception that is being controlled, and that
the same state of the same perception can be drawn from many different
combinations of lower-level perceptions, the multiplicity of lower-level
perceptions becomes, as Avery put it, a resource, not a problem.
-----------------------------------------------------------------------
Martin Taylor (950113.1630)--
However, a mild comment: What loop delays enforce is that control
cannot be maintained if the disturbance changes too fast. With
zero delay, control could, in principle, be perfect no matter how
fast the disturbance varied.
You would have to have zero delay, zero mass, and strictly proportional
functions (no integrators). In fact, a zero-delay system made of
otherwise real components would still have bandwidth limitations, and in
the case of human behavioral systems they would be scarcely
distinguishable from the limitations of the same systems containing
delays. You are making entirely too much of the effects of delays, and
forgetting that there are other considerations that have comparable or
greater effects on stability.
-------------------------------
RE: sleep study data
I can't find my key to your system for numbering the data files. I
unpacked the disk for 111.zip and obtained these results for all the
data files that end in A1;
File Track error model-real handle correlation
111001a1 0.047 0.9727
111037a1 0.039 0.9878
111073a1 0.025 0.9937
111109a1 0.043 0.9734
111145a1 0.026 0.9921
111181a1 0.028 0.9926
111217a1 0.034 0.9864
111253a1 0.082 0.9318
In the last one there was a single large reversal error and the tracking
looked pretty ragged. Are you sure you're reading your disturbance
tables in synch with the data tables? Check to see if you've compiled on
word or byte boundaries (I use word boundaries).
As I remember, A1 implies compensatory tracking and the lower
difficulty. So this should be one person's data for one week for one
experiment for one degree of difficulty. Doesn't look bad to me, but
maybe this first one was a fluke.
Could you check your analysis of these same files to see if we're
getting the same numbers? I used my copies of the disturbance tables and
they obviously are the same as yours WERE.
Also, when you present data it would help me if you would cite the file
name as above. No point in conjecturing about what's going on until
we're sure we're getting the same numbers!
----------------------------------------------------------------------
Best to all,
Bill P.