[From Bill Powers (921204.2100)]
Rick Marken (921204.1800) --
Read my algebra.
Rick, you have taken the most direct route to the correct answer while I
was still circling around. The error signal is not controlled. The way
you prove it is to add a disturbance to it and see whether the
disturbance is resisted. It is not resisted. Therefore the error signal
is not controlled: QED.
This gives me once again an opportunity to harp on HOW disturbances are
applied. You don't disturb a variable by just saying it has changed. You
have to establish a physical link in the model through which a
disturbing variable can contribute an effect to the disturbed variable
without FORCING the disturbed variable to change. If you FORCE the
disturbed variable to change by a known amount, you have broken the
control loop because you're preventing other influences from also having
their effects on the disturbed variable.
So, to explain to others, the way you inserted a disturbance of the
error signal was like this:
ref sig disturbance of error
> >
percep --> COMP --err sig--> ADDER --> effective error
> >
from input to output
function function
There is only one other way to disturb the error signal: add another
signal to the comparator. But that is equivalent to changing the
reference signal. Rick's approach is the only way to add a disturbance
to the error signal itself. And doing that CAUSES an error between
perception and reference.
ยทยทยท
-------------------------------------------------------------------
Best,
Bill P.
[Martin Taylor 921207 10:50]
(Olson, Cziko, Marken, Powers et al.)
If something is to be controlled, surely there must be a way for it to be
disturbed?
(Powers 921204 21:00)
You don't disturb a variable by just saying it has changed. You
have to establish a physical link in the model through which a
disturbing variable can contribute an effect to the disturbed variable
...
Where in the world (take the literal meaning of that expression) is the
possibility for disturbing the error in an ECS? It's a derived quantity,
based on the difference between a desired perceptual state and the
perceptual state that the world provides. That definition is changed by
any possibility of disturbing the "error" directly. A new quantity may be
produced, but it won't be the error any more.
To me it makes no sense to talk about controlling the error. One controls
things that could be disturbed. One brings the absolute error (the
difference between the controlled signal and its design value) toward zero.
That's just a way of saying that one controls the signal that can be
disturbed. In these discussions, that's the perceptual signal.
Martin
[From Rick Marken (921207.1400)]
Martin Taylor (921207 10:50) --
Where in the world (take the literal meaning of that expression) is the
possibility for disturbing the error in an ECS?
It could come from a stream of electronic pulses that is injected into
an efferent neuron that carries what is functionally an error signal.
Actually, I think you Canadians have a pretty good track record in
the area of disturbing neural signals (some of which were unquestionably
error signals) -- that Penfield fellow comes to mind.
It's a derived quantity,
based on the difference between a desired perceptual state and the
perceptual state that the world provides. That definition is changed by
any possibility of disturbing the "error" directly. A new quantity may be
produced, but it won't be the error any more.
What's your point here? Is a perceptual signal no longer a perceptual
signal if it can be disturbed directly (by injecting electronic impulses,
say)?
To me it makes no sense to talk about controlling the error. One controls
things that could be disturbed.
I guess you didn't like my proof. But your's doesn't seem like an
improvement (though it ends with the right conclusion). The problem
with your proof (really, it's an argument, not a proof) is that the error
signal can, unquestionably, be disturbed (see Bill's post that you
referred to -- Powers 921204 21:00). You might not like to call it an
error signal after that, but the fact that it can be disturbed suggests
(according to your argument above) that it is controlled. And that, as
Tricky once said, "would be wrong".
So I ask again -- what's your point?
Regards
Rick