[From Bruce Abbott (970104.1010 EST)]
Bill Powers (970103.0950 MST) --
You haven't been reading my posts closely enough. All I ask is that the
nature of the problem be identified to the point where the missing machinery
is specified in a plausible way. I believe I said something about at least
providing a box where the missing properties or functions are to go, with
some indication of the inputs and outputs of the box.
When you propose a model in which disturbances are simulated, you're
introducing functions that the basic PCT model doesn't have. You're adding
properties to the model without providing any machinery that would create
those properties. That's what I object to. If you were seriously proposing a
model, you would have to indicate how the effects you postulate come into
being, at least indicating that you're aware of the nature of the problems
that are raised by your proposal.
But I'm _not_ seriously proposing a model! I'm investigating how _any_
simulation-based model would behave, as compared to an equivalent control
model, under various assumptions about how well the "missing machinery"
would be able to do its job. That is all I intended to do with the RunBlind
simulation; it was created to serve a limited (but, I think, useful) purpose.
If you were to try to draw a _complete_ block diagram of the arrangement you
propose, even if you had to leave some boxes blank, I think you'd see what
I'm driving at. Every signal you propose has to come from somewhere, as a
function of other system variables or of independent variables affected by
something outside the system. If you look at any PCT or HPCT diagram, you
will find that this is always true: every signal is accounted for in one way
or another. No signal just appears by magic where it's needed in the system.
The point of the RunBlind modeling exercise is that it doesn't matter how
the regular disturbance pattern is extracted; whatever the process, it will
produce a simulated CV that can deviate from the intended CV in various
ways, and one can investigate how these various distortions would affect the
ability of the system to stabilize the intended CV under blind conditions.
That theoretical analysis will hold up no matter what physical device is
proposed for the job.
There are two issues here that keep getting conflated. The first is whether
anything useful can be learned from the theoretical comparison I provided.
We apparently differ on whether such an exercise is useful in the absence of
a detailed model whereby the simulated CV can be "extracted" from the input:
I say it is, you say it isn't. The second is whether RunBlind presents a
complete model; here you and I are in complete agreement that it is not (and
I said so explicitly when I presented it). However, you keep complaining
that it isn't, as if I had represented it as such.
Extracting repeating patterns of variation from an input signal seems no
more (or less) difficult a task than extracting objects and relationships
among objects from the patterns of rod and cone stimulation spread over the
surface of the retina. We don't have much idea at present how the brain
does either of these tasks. Lacking any real understanding of the physical
structures involved and their functional interconnections, I am in no
position to propose how such a system might accomplish its task of isolating
those repeating variations (Phase locking? Filtering?), but that shouldn't
stop me from examining how the performance of such a system would degrade
under various assumed faults.
I think I'm feeling resentful about the way you're playing this modeling
game. When I propose a model, really propose one instead of just
conjecturing to myself about various possibilities, I don't leave any loose
ends dangling. That's why I have to say, so often, "I don't know how to
model that." You're allowing yourself liberties that I don't allow myself.
You seem still not to understand the limited objectives of my efforts. I
was not proposing an alternative model, but only investigating how models
with the assumed capabilities would behave. As I have demonstrated, it is
entirely possible to make such a comparison without knowing how the
simulation-based system would extract its models.
I still like you, though.
Well, gosh, I still like you too.
By the way, have you had a chance to look at the SimFeed1 simulation I sent
your way?
Regards,
Bruce