[Martin Taylor 940614 11:20]
Rick Marken (940613.1400)
I realize that Rick often prefer magic to reason, so the following is not
very surprising:
Why "must" there be "ways in which control shifts from one perceptual signal
to another"?
The simple answer is that there are more independent perceptual signals
than there are independent mechanical actions with which we can affect the
world. By my estimate, which Bill Powers disputes, the ratio is on the
order of 10^4 to 10^6.
The phenomenal answer is that we can and do observe many things that we
do not at a particular moment control for, one way or another, though we
could and at some other moment may.
Do you
want a way for a control system to shift from controlling one perceptual
signal to controlling another?
I don't "want" such a thing. And your comment is ambiguous as to whether
you mean an ECS (which is suggested by your later text) or the hierarchy,
which is what I am talking about.
So there are two possible perceptual signals
entering the comparator and you want a switch that shifts from one to
another?
!!!
Or are you just saying that we must be able to shift
perceptual functions, maintaining the same physical perceptual signal path?
!!??!!
Or are you
saying we must be able to shift which of several lower level control systems
is actually used as a way of controlling some higher order variable (we
already have data the suggests that this kind of shifting IS done and it
is already handled, by adjustment of lower level reference signals, in the
plain vanilla HPCT model).
Probably, though not necessarily. The question in my mind, that leads me
to call it "alerting" is that the shift in which (any level) ECSs are
actively controlling their perceptual signals is occasioned by disturbances,
necessarily in perceptions that are not currently controlled, and usually
that are not subsequently controlled either. The changes occasioned by
changes in high-level references may or may not be of the same kind as
those that occur from alerting events.
Along with Bill Powers (940609.0910 MDT) I suggest that we get "back to
phenomena".
Well, I might ask the same in respect of discussions on the reorganizing
system. Clearly there are phenomena that we naively call "learning" or
"adaptation." Clearly Bill's reorganizing system is one way that such
phenomena could occur. But there is no necessity in phenomena that a
separate "reorganizing hierarchy" affecting the main hierarchy has to
exist or that if it does, it be the only mechanism for learning. But
the discussions tend to say that THERE IS a reorganizing system.
In talking about an "alerting system," I don't intend there to be necessarily
special mechanisms relating to the phenomenon of alerting, although the
different ways in which our senses work suggests that there is. I think
the evidence for this is at least as strong as the evidence for a separate
reorganizing system (as mechanism).
Same goes for the different characteristic levels of the hierarchy. They
may or may not exist outside of Bill Powers' fine intuition. I have learned
to trust Bill's intuition a long way, but I don't substitute it for
evidence. In the neural net literature, for example, it has been shown
that ANY configuration in time and space (i.e. any of Bill's perceptual
levels below category) can be learned by a network of nodes that consist
of a weighted summation followed by a nonlinear "squashing" function,
provided its input can include temporal history. So all the perceptual
functions in ECSs below category COULD be of the same kind. That doesn't
say that Bill P is wrong; it says only that his strict segregation into
intensity, transition, ... event, is unnecessary though convenient. It
could be an analyst's categorization based on the character of the CEV,
rather than a description of different types of PIF.
Phenomena, yes. But let's not discard physical limitations. You can't
control more degrees of freedom than are available in the narrowest part
of the combined set of control loops in the organism. I don't know where
the bottleneck is, but we do have a limit on the independently manipulable
joints, which puts an upper bound on it. And we have a limit on the
bandwidth of their actions. Any other factors can only lower this bound,
and make the problem of which perceptions to control from moment to
moment more severe.
Martin