🚀 go-pugleaf

[Martin Taylor 950304 18:00]

(Just dropped in to work to pick up some stuff for the trip)

>Bill Powers (950303.1720 MST) and later Bill Leach 950303.22:40 EST

>>Anyway, have you also considered this subject as it applies to
>>"information about ..."?
>
>    Yes, indeed.  I've tried to make it explicit, quite a few times.
>
>Better try again -- I don't recall any discussion of information in this
>particular context, which is knowing what information is "about" when
>you have only perceptual signals to work with.

Boy, do you shift the terms of debate kaleidoscopically!  My context is
that "perception of" and "information about" are both constructs that
relate to an outside observer's viewpoint.  "Perception" or "perceptual
signal" and "information in use" are aspects of what goes on within a
control system.

When you accused me of "cheating" and when Rick says that the control system
cannot "know" its output function or its environmental function, I have
always responded in exactly the same way as I now deal with "perception"
versus "perception of."  The outside observer can determine the "information
about;" the control system only uses it.

============
>If you have organized your perceptual systems so you perceive the
>category "rotten apple" on the basis of smell-signals, then you can
>experience "rotten apple" rather than inferring it. The debate, as far
>as my side is concerned, is not about the reality or veridicality of the
>referent* of any particular perceptual signal. It is about noticing
>whether there is any perceptual signal at all, or if you are creating it
>yourself independently of lower-level perceptions.

I thought we had got past that point several exchanges ago, in full agreement
with each other, and were going on to more interesting stuff.  (I assume
here that your last line is intended to read "independently, out of...").
From my side, we had agreed that if the latter is the case, then the
proper word is "inference."

>Be careful not to use the so-called "outside observer's view" as a way
>of cheating in the epistemological debate. There is no outside observer
>who knows what is really there when you perceive the apple smell or the
>category of appleness (ripe, green or rotten). The real outside observer
>is organized just like the observer in question, and has no better idea
>of what a "rotten apple" is in reality than any other person does.

Of course.  Is there, was there, any question of it?

=============
>     ... consider the flip-flop connection of perceptual functions, in
>     which the output of one PIF is connected with negative sign to the
>     input of another, and vice-versa.
>
>We have considered this type of connection several times. I agree that
>it probably exists, but it is only a special case. Not all categories
>require it, and in many cases a flip-flop connection would be a positive
>disadvantage.

Aha.  I knew I shouldn't have brought this up as I was on my way out of town.
There's too much to say on the topic.

No, I do not believe it is a special case.  Yes, I think the flip-flop
connection is what identifies a "category"-level perception and that
ALL categories require it.  Yes, to categorize is often a positive
disadvantage.  However, as you pointed out when I first brought this
up, the analogue data goes on by the flip flop, in parallel with the
flip-flop outputs, up to its own higher levels of perceptual control,
without being affected by the flip-flop action.

>It is also likely that the flip-flop function would be
>part of a logic-level system rather than a lower-level one

Yes, exactly.  It is what permits a logic-level system to BE a logic-level
system.  And all higher-level logical systems require as input the output
of a flip-flop or similar connection (or at least one that is likely to
be saturated).  But one of my points was that this logic can be developed out
of ANY of the analogue level systems, provided that the "real" world
(as encountered during control behaviour) is so constructed that the
implicated analogue signals are controlled in opposition--you can't increase
one without decreasing the other(s).

> particularly
>where the variables are sometimes treated independently and
>continuously, and at other times are treated as on-off logical variables
>with hysteresis.

Not sure what you mean here.  But one of the constructs (not sure whether
I mentioned it on this go-round) is a variable gain in the flip-flop
interconnection, just as a variable gain is a construct sometimes mentioned
as a possibility for the standard analogue hierarchy.  That produces a
set of perceptual signals that sometimes change continuously (but not
independently unless the cross-gain is zero), and sometimes discontinuously
and with hysteresis (but not as pure on-off variables unless the cross
gain is high enough to drive the perceptual signals into saturation).

>     This mechanism works for analogue perceptions at all levels of the
>     hierarchy, including, in particular, the relational level at which
>     we perceive things as being above, below, to the left of, inside...
>     other things.
>
>Unless you mean that this type of connection would perhaps sometimes be
>present at analog levels, I have to dispute your claim. The flip-flop or
>mutual positive feedback connection would be incompatible with analog
>control, because while in either stable state, the system would be
>totally insensitive to small disturbances, and would be unable to
>produce graded opposition to disturbances of arbitrary sizes.

You misinterpret.  The categories are valuable precisely BECAUSE they
are insensitive to small disturbances.  If you want a "red" thing, you
don't care about small variations in hue or saturation or brightness,
unless they make the thing "orange" or "purple" or some other category
that is not "red."  You control for the category, not the analogue
value of a particular specific red.

What I mean by "at all analogue levels" is that there is no reason why
a flip-flop group should not be driven off the intensities of analogue red,
green, blue... just as easily as by the analogue values of perceptions
"higher, lower, leftward..."  The analogue signals are still there for use
at higher analogue levels, unaffected by the fact that some of their
outputs go into other PIFs that happen to have cross-links.

I can't draw it well in ASCII, I don't think.  You did, a long time ago, so
I'll try.  The links from the red perception to the green PIF and from the
green perception to the red PIF have negative weight.  The "redness" and
"greenness" signals are intensity analogue perceptions.  The "red" and
"green" signals are intensity category perceptions, if you like that term
for category perceptions derived from intensity values.

    To higher               To higher
 "logical" levels        analogue levels

   |red     |green        |redness   |greenness
   |-      -|             |          |
   | \    / |             |          |
  PIF \  / PIF            |          |
  |  \_\/  / |            |          |
   \    \_/   \___________|_________ |
    \____________________ |         \|
                         \|          |
                          |          |


>I think you're letting a simple piece of technology run away with you. I
>can't see "It's all flip-flops!!!" as a catchy slogan.

No.  Catchy slogans aren't the objective of the exercise.  I am trying to
understand the relation between logical and analogue perceptions, as well
as a whole mess of other related effects that occur in reading and in the
reading disabilities that can occur.  And a lot else.

I agree that a flip-flop is a simple piece of technology.  So is an elementary
control unit.  That's one attractive aspect of it.  It does what seems to
need to be done to account for some of the things we observe (such as the
possibly universal occurrence of hysteresis in category perception), and
it meshes very nicely not only into simple association, but also with the
shift in the kinds of association that seem to happen as children age.

===================
>     I can perceive the opening of Beethoven's Fifth Symphony quite
>     well, merely by being exposed to those words (it happened just now)
>     with no orchestra or recording present.
>
>I hope you don't confuse this experience of the opening of the 5th with
>what you experience in a concert hall. What you're describing is
>imagination, not real-time perception.

Of course.  What else could it be?  But is not imagination ALSO "real-time
perception?"  My point was that it occurs without the sensory input that
would normally be required, but WITH the sensory input related to the
WORDS "Beethoven's Fifth Symphony."

Drat... there it goes again... Da Da Da Dum...

If I had written "The Sibelius Violin Concerto" I think I would have heard
something different.  And possibly vice-versa.

>An interesting point; we tend to say "inference" when we want to assert
>some degree of rightness or reality to what we imagine.

Maybe that's why the "reality" red-herring got into the discussion.

> "Inference", I suppose applies mostly in situations with a strong
>cognitive component. But it's still basically imagination.

I (now) think of "inference" as a process, the result of which is a perception,
as are the inputs to the process.  I think--and this demands discussion when
I get back--that the perception is a program-level perception.  I've never
been totally happy (though I get persuaded for hours at a time) with the
definitions, descriptions, and discussions of the program level over the
last three years.  I think that "the program level" might almost be
synonymous with "a cognitive component", but I'm happy to defer to better
judged opinion.  At any rate, it presumably requires categorized perceptions
as its inputs.  I'm guessing that the perceptual output of a program level
PIF might be an inference.  Is that total garbage?

>     There is a specific part of the hierarchy in which flip-flop
>     connections and associative connections occur.  The output of this
>     part is the category level of perception.  The "specific part" is
>     an interface between the analogue part of the hierarchy and a
>     discrete, logical, part of the hierarchy.  It is not a "level" as
>     such, in that the interface meets all levels of the analogue
>     hierarchy.
>
>That word "meets" is sufficiently vague to cover just about anything we
>find, isn't it?

I don't like the word, either, but I couldn't think of a better.

What I mean is that the signals that branch off to become inputs to
cross-linked PIFs could be signals at any level of the hierarchy.  I'm not
intending any metaphor.

>It also
>agrees with my own general, but not very well-developed, idea that the
>input signals for any level can come from any lower-level system.

Perhaps, but this is very specific, in that the suggestion applies only
to signals from the same level going to closely coupled PIFs of a specific
kind.  In other words, as I have, I think, once drawn in ASCII, I put
the "category level" beside the whole analogue hierarchy, as the base
of a parallel logical hierarchy, rather than as level N of a single
hierarchy.  This organization allows for analogue versions of any logical
signal, and neither analogue nor logical has primacy.  A political action
can be perceived as "autocratic" (therefore not "democratic") and at the
same time have an analogue value of 2 on "democraticness" and of 7 on
"autocraticness".  The outputs of ECUs in the logical part might well
contribute to reference signals in the analogue part, but I think not
vice-versa (which might lead to a possibility for some experimental
predictions that I don't now see).

========================================
========================================
>Bill Leach 950303.22:40 EST

>I did
>not however take him to mean 'flip/flop' in these sense of a good digital
>logic designer using ECL stuff.

But that's what I did mean.  The difference is that in ECL terms the cross
gain is high and the two outputs saturate except during the very brief
transition between states.  As I conceive the neural flip-flops, the cross
gain is normally low, and the loop cross gain may even sometimes be less
than unity.  The transitions of states, even with high gain, may not seem
instantaneous.

>I admit that my statement is pretty loose... how would we 'recognize'
>that a mutually exclusive condition exists?  Particularly, without and
>XOR 'gate' of somekind?

There are two kinds of mutual exclusion to worry about.  One is that the
world does not provide both possible perceptions--as it happens.  Perhaps
one COULD have a patch that is both a good red and a good green at the
same time, but we have never encountered it.  The other kind is that
the perceptual system doesn't allow us to perceive both possibilities
at the same time.  A reversing figure is a good example of this situation.
In the "old woman-young woman" figure, both heads are implicit in the
drawing, but we have a very hard time trying to see them both at once
(I've never managed it, after a lot of trying).

Typical experiments are like the ones that Bruce Nevin has been talking
about--gradually changing some acoustic parameter and determining when
the listener hears something different.  Going from /b/ to /d/ doesn't
give the same results as going from /d/ to /b/.  There's hysteresis.

======================
Degrees of freedom:

>Since Martin will be gone for a week, maybe I can 'slip' this one in. :-)

Sorry, old chap!

>Is it not possible that an individual might have several million
>references ABOVE the intensity level?  All as active control loops with
>'nothing going on' (ie:  the reference condition is satisfied)?

Possible, yes.  Likely, yes.  If PCT is to be believed, almost surely yes.

The DF problem is with ANY bottleneck anywhere in the multitude of control
loops through the outer world.

>The 'perceived' alerting then could well be caused by a perception that
>causes an error in one of these normally satisfied systems.

Yes, I assume so.  But note that it is not necessarily true that the
control behaviour affects the perception that corresponds to the alert.
It is often true, but not necessarily so.  The alerting perception may
have a built-in reference of zero, which would mean, in effect, that the
perceptual signal is simply passed on, to become part of a perception
elsewhere, and THAT perception might then differ from its reference and
cause overt output.  Anything of that kind is possible (see below).

>It seems likely to me that error signals 'compete' for resources (degrees
>of freedom of output issue) and that there must be an arbitration
>mechanism present

Yes.  One bottleneck is in the joints and musculature.  There may be
thinner ones.

>I know that this is sort of a restatement
>of one of the things that Bill P. and Bruce A. have been talking about.

In an earlier incarnation (more than one) of the DF discussion, Bill P
argued that disturbances in the world were slow enough to allow the living
organism to time-multiplex the output.  I agree totally.  That's the basis
on which I predicated the notion of alerting as being a consequence of PCT.
You are putting it in much the same way as we all were, even when we seemed
to be contradicting each other.

>I like to think that 'digital behaviour' is digital only in appearance.
>Thus, for this competing for resources idea is it not reasonable to think
>that maybe error signals inhibit each other on a priority basis via
>non-linear analog gates such that a weak high priority error will
>override a weak low priority error but not a strong one and a strong high
>priority error overrides all low priority errors?  Sort of like biased
>auctioneering?

Yes, we discussed this as a possible mechanism.  It's quite plausible.  And
there may be a whole variety of mechanisms operating in parallel, of which
this is one.  Mechanism is an issue separate from the need for an alerting
effect, which you discussed above.

>Of course this sort of explaination offers a terrible dilemma for the
>researcher since it means that there are many 'wants' that are not only
>unknown but as a practical matter are also unknowlable.

I think this is true.   But I'd go further, and suggest that you yourself
may have many wants that are unknown to you, and may never become known
to you unless an appropriate disturbance alerts you to them. Or, in case
Bill P thinks I have forgotten, unless appropriate changes in mid-level
reference signals induce conflict in that part of the hierarchy.

Now I really will be gone for a week.

Sorry, Bills.

Martin