Bjorn Simonsen
(2007.10.26,09:35 EUST)–
But back to
your comment above. r10 must come from neurons “above”
the p10, r10 level.
It could, I suppose. But I was
thinking that it was more likely to be
a property of the cell body itself. The cell body could act like a
semi-conductor, only producing output (e) if the input (p) is
greater
than some amount (functionally equivalent to r). So e = p - r for r
and p> 0. The value of r is a property of the cell; and it defines
the
upper limit of p that produces no error. In real nervous systems all
signals are positive (assuming they are firing rates) so there would
have to be another cell that takes an inverted p as input with the
same r and then the outputs of both would have to be combined
appropriately to get the net error, I think. Bill Powers knows how
to
build such circuits. But my point is that, since r for the highest
level systems isn’t varied by a still higher level system to achieve
that system’s goals, it can be a relatively constant property of the
cells involved in the comparator process, being changed only over
the
long run by random reorganization.
This reference, r10, could be built into the cell body of the
neuron
doing the highest level comparison; these is no need for an
actual
reference neuron to set r10 since r10 is not “used”
(i.e… varied) by a
higher level system to achieve a higher level goal.
I don’t fall for your explanation that r10 could be built into the
cell body
of the neuron at the real 11 th level. That is because I think
comparators
at the 10 th level can accept different r10 (plural) at different
time. I
can’t understand your certainty that r10 is not varied by the
neurons at
the real 11 th. level
I think it’s more a matter of modeling rather than of certainty. If
the 11 level is the highest then all it is is references. There is
no
higher level reason for those references to be varied to I would
model
them (neurologically) as being part of the neural comparator that
transforms level 10 perceptions into level 10 errors. The level 11
references are the constant “offset” of this transfer function
and can
be implemented as a fixed property of the neural comparator; there
is
no need to have a neural signal carrying r; you could have such a
signal, but why? The data suggest that people rarely vary their
references for their highest level perceptions over the course of an
entire lifetime. How frequently do you change your religious goals
over the course of a year? I know that mine haven’t changed much
since
I was about 5. As they say, once a secular humanist, always a
secular
humanist;-)
In fact, I find the current mania for molecular biology that I
see
in psychology a rather pathetic sign of how stuck in the dead
end of
lineal causal thinking current psychological modeling
is.
May I ask in another way. How do you model what you say about
"it could
just exist as the offset in the highest level
comparator"?
I hope I gave a better approximation to how I would do it in my
first
paragraph above.
I thought the perceptions at the (real) 11 th. level are based
on
combinations of principle level perceptions. If that is correct it
can’t
be constant. Where am I wrong?
If there a perceptions controlled at the 11th level then it is the
12th level references for those perceptions that I am talking about.
If the system level perceptions are 11th level then the references
for
those perceptions are what I would call the fixed 12th level
references. It’s the top level references – regardless of the
number
of levels in the hierarchy – that I would say are likely to be
fixed
properties of cortical neurons.
Yes, I guess all r (plural) are set by reorganization.
No, I think only the top level references are set (and changed) by
reorganization. All other references are varied as the by higher
level
systems as the means of controlling their perceptions.
I read your text as if r10 is a reference for a certain principle
(10 th
level). But I can’t understand why it can’t be varied. A US Army
soldier
wish to do different things if he fights in Iraq or he marching on
Broadway
in the Eastern Parade. In both situations he wishes to behave as a
US Army
Soldier.
Don’t get so exact about level numbering. Yes, people vary their
references for principles to control higher level goals. If those
higher levels goals are at the top of the hierarchy (say they are
system concepts) then I am saying that the references for those
perceptions, which would come from a next level up, don’t actually
come from a next level up; they are just an offset in the comparator
at that higher level.
Best regards
Rick
–
Richard S. Marken PhD
rsmarken@gmail.com
–
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.5.503 / Virus Database: 269.15.11/1094 - Release Date:
10/26/2007 8:50 AM
–
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.5.503 / Virus Database: 269.15.11/1094 - Release Date:
10/26/2007 8:50 AM
]From Bill Powers (2007.10.0430 MDT)]
Rick Marken (2007.10.26.1700) –
I’m confused by this contraction of the hierarchy. What level are you
both leaving out? I have always considered that there are 11 levels, the
11th level being complete with an input function that perceives a
collection of principles to produce a perception of the state of a system
concept, and an output function that converts an error in system concept
into a set of reference signals specifying various principles. I would
have thought that the problem is the source of the 11th level reference
signals, which can’t come from any higher systems (unless, as is
perfectly possible, there is at least a 12th level of control that is so
much taken for granted that it’s so far invisible).
“The comparator” at any level is a reification of the function
of comparison, and comparison is just one way of viewing the fact that it
takes a nonzero amount of perceptual input to produce zero output action
(or put differently, that zero input produces non-zero output). What I
refer to as comparison might take place in a specialized comparator cell
or network, but it could also take place in an input function or an
output function if the reference signal went there instead, with suitable
excitatory or inhibitory connections.
As Rick suggests, and as I said or implied in B:CP (I believe), any
property of neural connections that produces (and possibly varies) the
offset mentioned above qualifies as a reference signal, even though there
may be no explicit neural signal carrying the reference information. If
the threshold of response of a neuron to another neural signal is 10
impulses per second (that is, the output frequency does not begin to rise
above zero until the input frequency is at or above 10 impulses per
second), there is an effective reference signal of -10 impulses per
second. If this threshold is affected by the concentration of circulating
chemicals, then that contribution to the net reference signal is
chemical, not neural.
We should also not forget that a magnitude of zero is a perfectly good
magnitude for a reference signal. If the reference signal is zero, then
any amount of perceptual signal will produce an error signal (but, of
course, only if the reference signal is inhibitory and the perceptual
signal is excitatory). The error will be zero if and only if the ensuing
actions at lower levels succeed in bringing the perceptual signal close
to zero. This is what we call “avoidance” behavior. To avoid a
perception is to try to bring its magnitude to zero. When any new level
of control is just starting to be organized, and for the time being is
the highest level, we might expect the resulting control behavior to be,
at first, avoidance of the newly-formed perception. This might also be
the case for perceptions that appear unexpectedly, even if one normally
sets high reference levels for them.
Let’s think about inhibitory and excitatory signals for a
moment.
Suppose the INHIBITORY reference signal has a magnitude of 100. Then
excitatory perceptual signals must have a magnitude of at least 100 units
to start producing an error signal. There will be no behavior until the
perceptual signal has reached 100 units of magnitude. A perceptual signal
of 110 units will produce 10 units of error signal, and that will result
in behavior tending strongly to bring the perception down toward a
magnitude of 100.
To an observer, it will appear that this system simply doesn’t detect
anything less than some minimum amount of the stimulus, and that it
responds to any larger amount by producing some action. If the proximal
effect of the stimulus isn’t known or observed, it will not be apparent
that the effect of the action is to keep the effect of the stimulus on a
perception from getting any larger than the threshold amount.
So we can discover one-way control systems with excitatory perceptual
signals and inhibitory reference signals by looking for “thresholds
of detection.” The fact that there is a threshold shown by the
response to the stimulus does not mean that there is no perceptual signal
below that amount of stimulus. It just means that there is some nonzero
amount of inhibitory reference signal that has to be exceeded by the
excitatory perceptual signal to result in error, and thus action. This
may have a considerable significance for phenomena such as
“blindsight.”
The other side of the coin is one-way control systems with inhibitory
perceptual signals and excitatory reference signals, more like the kind
we normally talk about. If the perceptual signal exceeds the reference
signal in magnitude by any amount, the action will just be zero. There
will be action if and only if the perceptual signal is less in magnitude
than the reference signal. If the reference signal is positive and the
perceptual signal is zero, the action will quickly raise the perceptual
signal to match the reference signal. So one-way control systems of this
kind are typified by what seem to be voluntary actions which cause
perceptions to arise that would not otherwise have appeared. If large
magnitudes of the same perception are caused externally, there will be no
action: the same system with a low reference signal will ignore
them.
I now leave it as an exercise for the student to deduce what will be
observed when there are two systems controlling perceptions that depend
on the same environmental variable,with one system being of the first
type and the other system being of the second type. What will happen when
the two reference signals specify the same level of the controlled
variable, and what will happen if the two reference signals differ by a
substantial amount in either of the two possible directions (A larger
than B, or B larger than A). Qualitative answers will receive a D, and
quantitative answers will be graded on correctness (by someone
else).
Best,
Bill P.