[From Rick Marken (931116.1300)]
Martin Taylor (931116 12:00) --
I said:
Unless you know of a dynamical system that resists disturbances
we're looking at systems that DON'T CONTROL.
Martin says:
Tautology.
Aw. Can't I have a "sophistry" like you gave Bill. I'd even accept
the "simple sophistry" that Bill spurned.
A control system is the type of dynamical system that resists
disturbances. There are others. A marble in a bowl resists disturbances
to its resting position in the bottom of the bowl, but is not a control
system.
I can't believe you claim this? What kind of marbles have they got
up there in Canada? So when you gently push on a marble at the bottom
of a bowl the marble pushes back with a force that is nearly equal and
opposite to the applied force -- resulting in a stationary marble?
I'd love to see THAT! Is it in the design of the marble or the bowl?
I said:
So, given the now agreed on terminology,
I think we can see that there is NO information in perception about
disturbances but there IS information in perception about fluctuations (in
the CEV).
You say:
I still can't agree with this.
Too bad. A-diverging we shall go.
Perception carries information about the CEV, and THEREFORE about the
effect on the CEV of everything that influences the CEV, whether it be
a disturbance or the ECS output. The question as to whether the different
effects can be segregated has no bearing on the question of how much
mutual information there is.
So we are back to "there IS information about the disturbance in
perception", a claim that you have no way of substantiating except
by saying "perception carries information about the CEV, and THEREFORE
about the effect of ... a disturbance". Now THAT sounds like a
tautology. Actually, it sounds like the desperate cry of one
unwilling to abandon a long-held assumption that is demonstrably
false -- the assumption that perceptual information guides the
outputs that produce intended results.
I can't seem to get you to see things from a PCT perspective -- so
enjoy your assumptions. I guess I'll just wait for your paper
describing how a control system gets the information about
disturbances out of its perceptual signal. Actually, Tom Bourbon
(931116.1053) just described the only two ways I can imagine this
being done. Unfortunately, Tom shows that the information
about the disturbance is not obtained from the perceptual signal
representing the CEV that is being disturbed (at least, not from
that perceptual signal ALONE); the disturbance can be inferred
from the perceptions in several control systems.
Hans Blom (931116) --
In my view, feedforward has an important place in
control, and is most fruitfully combined with feedback control.
You make this claim even though you have presented NO evidence (other
than the notorius "walk in the dark" anecdotes) that feedforward is
needed to explain any aspect of human behavior.
My emphasis on feedforward must not
be construed as a dislike of feedback.
It's being construed as an attempt to provide an explanation for
data that don't exist -- ie. behavioral data that cannot be
explained by a feedback control model. This makes it SEEM like
you have an agenda (regarding feedforward) rather than a purely
scientific interest in how living systems work. If you point us
to the data that seems to demand incorporation of feedforward
mechanisms for their explanation, then we will be more interested
in figuring out how feedforward might fit into PCT.
My emphasis on feedforward does not go back to "outdated"
psychological paradigms, but has to do with control engineering themes
like "graceful degradation" in the face of sensor malfunction, "inher-
ently safe" controllers and the like. Are these issues that would have
been unimportant in biological evolution?
Who knows? But we (PCTers, anyway) don't do science by asking whether
something would be considered important to biological evolution. If you
want to use feedforward as a tool in your engineering applications that's
fine. But don't assume that successful application of feedforward in
engineering means that feedforward MUST be a part of models of
living systems. Wait for the behavioral data.
Actually, "not waiting for the data" seems to be a VERY common problem
in the social sciences these days and it may be one reason for the
growth of "trendy science". In the old days, scientists made observations
and then tried to think up models to explain them. They then tested
these models by making further obsevations. In the new, trendy science
days, scientists seem to depend FAR less on observation; instead,
they look at all the whiz bang toys produced by the engineers -- toys
that often SEEM to behave like "real live" living organisms -- and
assume that the mechanisms used to build such life-like devices
MUST be the basis of life. So we get dynamical systems, chaos, fuzzy
logic, expert systems, etc. proposed as underlying mechanisms of behavior,
not because they precisely account for any behavioral data but because
they produce the APPEARANCE of life-like behavior.
PCT starts with PHENOMENA, not models. We observe control in simple
situations and develop models that account nearly perfectly for control
in these situations before moving on to more complex behavior. We know
the PHENOMENON that we want to explain (purposeful behavior) and we
have a way of determining whether we are looking at an example of that
phenomenon (the test) when we are looking at behavior. If someone
builds a device that looks like Abe Lincoln and moves its lips while
delivering the Gettysberg address, we don't rush out and claim that
the mechanisms that produce this behavior must be part of an explanation
of how people speak. FIRST we determine whether the device is controlling
any perceptions; then we see if the behavior we observe is a side effect
of that controlling. Then we test people to see if they control the same
kinds of variables that the model controls. If it turns out that they
do, then we start talking about the device as a possible model of the
mechanisms that underly behaviors like giving the Gettysburg Address.
I'm glad you've had success using feedforward in the control systems
you've built, Hans. But we're not trying to build a better control
system; we're trying to build as simple a model as possible that
will capture the OBSERVED complexities of the controlling done by
living systems. We'll put in feedfoward if we need it but currently,
there is nothing (other than the say-so of you, Martin and, now Oded)
that seems to demand it. I am currently working on an experiment
to see if I can get data that DO demand the addition of some kind of
feedforward to the PCT model. I really hope that it comes out "right"
(requiring the feedforward) just to prove to you feedforward types that
we PCTers have nothing personal against feedforward, per se.
Best
Rick