Trying to understand

[From Bill Powers (950105.0715 MST)]

Martin Taylor (950104.0600) --

     But otherwise, this get odder and odder.

     What makes you say the following?:

     > You want to "simplify" the process of reorganization by
     >eliminating the capacity to alter the hierarchy structurally,

     Absolutely not. I don't. I DON'T, i dOn'T!!!

     The business of reorganization is structural alteration in the
     hierarchy.

What makes me say it is the following:

     Martin Taylor 950103 11:30

     In order to make the proposal as straightforward as possible, I
     suggested a reorganizing system in which everything was the same as
     in the "standard" system except for its localization. I PROPOSED
     THAT THE OUTPUTS OF THE CONTROL SYSTEMS FOR THE INTRINSIC VARIABLES
     CONTRIBUTE TO REFERENCE SIGNALS IN THE MAIN HIERARCHY, RATHER THAN
     CAUSING CHANGES IN THE STRUCTURE OF THE MAIN HIERARCHY. This seems
     to me to be reducing the number of ad-hoc processes, not increasing
     them. [Capitalization added]

After reading your present post, I think I finally get the idea you're
proposing. You are saying

1. Errors between intrinsic variables and inherited reference levels are
not the cause of reorganization. (I had not understood that you were
abandoning the idea that intrinsic error drives reorganization)

2. The system for controlling intrinsic variables is organized exactly
as in B:CP, except that the output signals coming out of the
reorganizing system (so-called by me) and reaching all levels in the
hierarchy, as shown in Fig. 14-1, p. 188, are reference signals rather
than effects that cause structural reorganizations. This is suggested by
your post of 950104.1010:

     What occurs to me now is that the localized reorganization
     principle allows the intrinsic variable control systems to send
     output that contributes to the reference signals at ANY level of
     the main perceptual control hierarchy, even very low down.

3. All structural reorganizations are caused locally, by a reorganizing
aspect of each ECU, and are based on the error signal in the local
control system (as in Tom Bourbon's "adaptive control" model).

Before we go on, please verify that I am understanding you correctly.
Also, I suggest that you re-read B:CP Ch. 14.

···

----------------------------------------------------------------------
Best,

Bill P.

[Martin Taylor 950105 16:30]

Bill Powers (950105.0715 MST)

On what the localized reorganization proposal is:

After reading your present post, I think I finally get the idea you're
proposing.

I suspect you are right.

1. Errors between intrinsic variables and inherited reference levels are
not the cause of reorganization. (I had not understood that you were
abandoning the idea that intrinsic error drives reorganization)

Not the DIRECT cause, no. But since the proposal takes ANY sustained
and (especially) increasing error as the driving effect for local
reorganization, intrinsic error ALSO will drive reorganization of the
linkages of its ECU's output signal to other reference inputs in the
hierarchy. So in this proposal, the control of intrinsic variables
is ordinary control, but if that fails, then reorganization happens.

If the rest of the hierarchy of perceptual control is controlling perfectly,
but the intrinsic variable control isn't, then the local reorganziation
will provide different reference signals to possibly different ECUs in
the rest of the hierarchy, in common language "getting the organism to
try something different that it already knows how to do."

But if the failure of control of the intrinsic variable is affected by
failure of control in some supporting part of the hierarchy, then that
part of the hierarchy ALSO will be affected by reorganization.

2. The system for controlling intrinsic variables is organized exactly
as in B:CP, except that the output signals coming out of the
reorganizing system (so-called by me) and reaching all levels in the
hierarchy, as shown in Fig. 14-1, p. 188, are reference signals rather
than effects that cause structural reorganizations.

Yes, apart from changing "all" to "any" in the third line. I think your
criticism on other occasions of "level-jumping" output signals probably
applies here. If the localized reorganization proposal works at all, one
result should be that the outputs of the intrinsic variable ECUs provide
reference signals to those places in the hierarchy where they do the most
good, and most redundant or counterproductive linkages should be eliminated
over time, by reorganization.

3. All structural reorganizations are caused locally, by a reorganizing
aspect of each ECU, and are based on the error signal in the local
control system (as in Tom Bourbon's "adaptive control" model).

Yes. His was a good model for what is going on.

Also, I suggest that you re-read B:CP Ch. 14.

I'm obviously going to have to either break down and get another copy, or
get our library to buy one. I've been totally unsuccessful in finding out
where my copy has got to. I read it through once, and parts of it more
times, lent it out, and it has vanished. I have held on to the hope that
it would return, but that hope has been dwindling over time.

Anyway, I think we are now at a good restarting point for a critique of
the localized reorganization proposal. Once the nature of it is understood,
I'm quite happy to participate in destroying it if it won't stand up.

One place that may fall quite easily is the GA proposal for making new
ECUs, but I'm not surrendering it right away! And it isn't intended to
replace the idea that there is an initial randomly connected mush of
ECUs (or neurons) within which inappropriate linkages are deleted and
effective ones enhanced as part of the reorganization process--or any
other related notion. It's an additional suggestion.

Martin