( Gavin
Ritz 2008.12.30.11.40NZT)
[From Bill Powers
(2008.12.29.0940 MST)]
Gavin Ritz 2008.12.29.16.36NZT
See my comment far below.
Thank you for your frank response to my
conundrum. PCT doesn’t have the answer and seems that it’s your
conundrum too.
This makes no sense at all. What specifically is a neural function?
There is one page on this in Behavior: the Control of perception (Premise about
Brain function) and it says very little about what this is and no definitions
in the same book under the definitions section at the end. And it’s not mentioned in any of
the other books as far as I can tell. Not easy to navigate the books because
most of them don’t have an index.
There’s a glossary in the back of B:CP that might help,
It’s not in the glossary at the
back. Unless it’s worded differently. There is no explaintion of neural
function. And it’s not explained anywhere else.
but I can see that the
explanation has to go a little deeper. If I get too elementary don’t take it as
an insult – I’m just guessing where to start. When I wrote B:CP I assumed that
I was writing for people who already knew these things. That was a mistake, I
discovered.
First, what PCT is.
PCT is basically a model of the brain. It’s meant to explain how we experience
things and act on them. The basic assumption is that both experience and action
are closely linked to activities in the brain. As far as I could go in the
years between 1953 and 1972 when I submitted B:CP for publication, I studied
(every now and then) how the brain works by reading books on neurology and
brain functions. There wasn’t a lot available that was useful in understanding
how behavior works, but at least I learned the broad picture of how the senses
work and how neural signals in the brain get turned into the motor actions we
call behavior. I don’t mean I became an expert. But where I had to guess, it
seems that I didn’t go too far wrong.
Perception
We experience the world through our sensory receptors in the eyes, ears,
fingers, nose, mouth, gut, and so on. If the neural connections carrying any
kind of sensory information are damaged, that part of the world disappears from
experience, so we know that experience depends on the existence of those
connections. We experience only what our senses tell us of the outside world;
if there are things out there that don’t affect our senses we don’t experience
them (like ultraviolet light, x-rays, and other things we have detected with
artificial sensors – or have deduced with logic, like electrons or gravity).
The term “signal” is used in PCT as it is used in electronics. It
does not mean the same thing as a traffic signal or a signal to a waiter or a
pistol shot that starts a row of runners going. It means a train of neural
impulses being generated by a nerve cell and traveling to the input synapses of
one or more other nerve cells, or to a gland or a muscle fiber. In electronic
systems, signals are carried by wires; in the brain, by nerve fibers. Some are
carried by chemical concentrations, but that’s in the category of “further
information.”
A perceptual signal at the first level of organization in the nervou s
system carries information about how much stimulation is currently acting on a
sensory receptor. If the stimulus is weak, such as a faint light intensity
reaching a rod or cone cell in the eye, the signal consists of impulses
occurring slowly, say 5 or 10 times per second. As the light intensity
increases, the impulses occur more and more rapidly, and for really bright
light may reach frequencies of 500 impulses per second or more. The rate at
which impulses are generated thus is a measure of the intensity of light
falling on the sensory receptor cell (really, some small group of cells, but I
don’t want to complicate this). The same kind of relationship between stimulus
intensity and frequency of impulses generated by a sensory cell occurs for all
forms of sensory receptors. All that we, as brains, can know about the world
must be contained in the set of all sensory signals coming inward from sensory
receptors.
Functions
There are many levels of organization in the brain. The sensory signals leaving
the receptors reach neurons in the brainstem, which send new signals to the
midbrain, and so on layer by layer to the cerebral cortex. At each level, the
receiving cells receive signals from more than one source, and emit a single
signal that goes on to further layers (and which also follows lateral pathways
to other neurons at the same level which we will get to).
When multiple signals are received by a single neuron, they affect voltages
internal to the cell – actually, the concentrations of positively and
negatively charged molecules, which interact with each other both electrically
and chemically. When the cell-wall voltage exceeds a threshold, the cell fires,
discharging it and generating an outgoing impulse, racing away along the axon,
the long neural fiber that carries the output signal. Ion pumps then recharge
the cell wall’s voltage. The threshold voltage inside the cell, affected by the
incoming impulses, determines how rapidly the cell will generate outgoing
impulses. The result is that the frequency of the outgoing train of impulses
depends on the frequencies of all the input signals, not just one of them, and
in a complex way. I laid out some of the ways in chapter 3 of B:CP, including
ways in which signals can be generated by several nerve-cells acting on each
other.
Neural computing functions are composed of multiple nerve cells that send
signals to each other. At each level in the brain, we find not just a
single layer of cells relaying signals to higher levels, but what are called
“sensory nuclei”, masses of millions of cells with complex
interconnections. Out of these nuclei come hundreds or even hundreds of
thousands of signals going upward toward the next level, and sideways to other nuclei,
and each one of these signals has a momentary magnitude (that is, frequency)
that depends on the magnitudes of many of the signals entering the nucleus from
below.
The magnitude of each outgoing signal, if I may switch from frequency to
magnitude without creating objections, depends on the magnitudes of some number
of signals entering the nucleus, input signals which are the output signals
from levels below. Here is where the term “function” enters in its
mathematical sense. If y is the magnitude of an outgoing signal, and x[i] is
the magnitude of the i-th signal in an array of input signals, we can say that
y = f(x[1], x[2], x[3] … x[n])
where n is the number of input signals affecting the output signal y. This
doesn’t tell us the specific formula that involves all of the x’s; it just says
there is some formula, probably a different one for every output signal y, and
it tells us which inputs are involved in generating the output signal
represented by y.
The letter f means “function.” We use it to refer both to the
physical set of neurons involved, and to the mathematical representation of
their actions and interactions. Mathematically, functions are expressions,
formulas, in which the listed variables, the x’s, appear. When the magnitudes
or values of all the x’s are given, the function can be evaluated (once the
formula is known) to compute the magnitude of y, the output signal. In this way
we pass from the physical, neurological description of the networks of nerves
to mathematical expressions describing how each output signal depends on
different sets of input signals. As a shorthand for this we simply say that y
is a function of the set of x’s. We also say that the physical neurons
“are” a function. We mean by that that we have a mathematically
defined function which more or less adequately defines the way in which the
output signal in axon y depends on the input signals coming from input signals
x1 to xn. And when we say that one signal depends on another signal, we mean
that the magnitude of one signal
depends on the magnitude – not
just the mere presence or existence – of another signal. We measure the
magnitudes of neural signals by measuring their frequencies.
The whole trick in understanding perception is that of defining the functions
that connect one layer of sense-based neural signals to the next layer up. Here
the books fail us. Nobody knows. Discovering the forms of these functions is a
very hard problem that has not yet been even partially solved. So, are we
stuck?
Yes I’m stuck here too, maybe I
expected too much from PCT. And I’m asking Rick to explain this and he
can’t.
No, because each of us has an instrument that can show us something
about the nature of all these neural functions. All we have to do is look at
the world around us and inside us. Since everything we experience has to start
as intensity signals coming fron sensory endings, we know that what we are
experiencing must be a very large collection of neural signals. These signals depend on physical interactions in the
external world, but they are not the same thing as the external world, and they
are not in the external world,
either. They are in our brains.
In an appendix of Making Sense of Behavior,
I outlined what I think I have found out about 11 levels of perception. Each
level consists of a set of neural functions (the forms of which we do not know)
which generate neural signals which we experience as various aspects of the
world and ourselves. What I found was that we can define types of perception,
one type per level, such that a perception of a higher level or type is a
function of some set of perceptions of a lower level or type. A configuration
– a shape for example – is composed of perceptions that are not
configurations. We call them sensations. Without any sensations, no
configuration can exist, but sensations can exist without also being seen as
configurations. This tells us which type depends on, is a function of, which.
We still don’t know the form of the function, but we now know where to put it:
between the signals indicating sensations, and the signals indicating
configurations. And we know the direction of the transformation: from
sensations to configurations, not the other way around. In the context of what
remains unknown, that isn’t much, but compared to what we knew before, it’s a
lot.
All 11 types I found are related in this way. They also satisfy other constraints
having to do with controlling perceptions at each level. I’m not sure I have
defined all 11 correctly, or that 11 is the right number, but I think this
picture puts us on the right track. It took me about 40 years to come up with
these 11 levels, with the last rearrangement and expansion having been
suggested by other members of the Control Systems Group in the 1990s.
Reference signals
Control depends on sensing the current state of some variable, comparing it to
a desired state, and using the difference as the basis for generating actions
which will change the current state so it is closer to the desired state. We
must therefore ask how the desired state, which seems a purely mental concept,
can be compared with the current state, which seems purely physical.
I brought that up partly to show how irrelevant the distinction between mental
and physical is in PCT. Experience, which is mental, is experience of neural
signals, which are physical. The two realms are one. The nature of awareness,
which is involved in conscious experience, doesn’t have to be explained right
now, because all we’re working on is what there is to be aware of. And that, in
PCT, is the world of neural signals.
If the current state of something being controlled is represented as a neural
signal, it would seem to make sense to say that the intended state of that
something must also be represented by a neural signal. Inhibitory neural signals
subtract from excitatory neural signals, so the output of the cell doing the
subtracting indicates the difference in magnitudes of the two signals. We call
that a comparator function, and the signal it emits the error signal.
The error signal
represents the difference between what we want and what we’ve got.
Well so does the implication of a goal.
We don’t know right now
where the reference signal comes from, but we’re about to.
This is what has been bugging me, because in
the books not sure which one it says that this reference signal is an intention
or goal. This cannot be because then you assume another state (two states). Any
goal or intention requires two states. Intention the word comes from the Latin
to stretch. A goal too must have a desired state and one that one is in lets
call that the “ground state”. So what has confounded me is the
reference signal is like an error signal too.
Looks like I’m stuck here and so are
you.
Thank you for your frank response . basically
then PCT doesn’t have the answer to the source of the reference signal.
I’ve got the rest pretty clearly actually.
Regards
Gavin
The error signal is routed, through more neural functions that we can’t
describe in any detail, to lower-level systems. To what part of the lower-level
systems? To the comparators. The output signals from the higher system are
wired to act as reference signals for the lower systems they reach. They don’t
tell the lower systems what to do, what actions to perform. They tell each
lower system how much of the perception it senses to want.
Every level but the lowest acts in that way – not by producing behaviors, but
by specifying the amount or level of perception that lower systems are to
produce by varying their actions. Each level can cancel the effects of
disturbances on the perceptions controlled at that level without being told to
do it, or how to do it, or when to do it. Only the lowest level accomplishes
its control of sensed muscle force by sending its output signals to muscle
fibers.
Conclusion
So there is Hierarchical Perceptual Control Theory. Notice that we never talk
about what behaviors people will produce under what environmental conditions or
as a result of which events. The point of PCT is to explain how it is that
people can behave at all, and behave so as to create predefined consequences.
We aren’t concerned with which goals people seek or why they seek one goal
rather than another. We want to know what a goal IS, that it can influence
behavior in any way. We want to know how the control of one set of perceptions
comes to be a means of controlling some other perception.
In short, PCT is an attempt to explain how
behavior works. Any behavior, by any organism, anywhere, any time,
of any kind. PCT replaces the simple notion that organisms are made to act by stimuli.
It puts a scientific footing under psychology and builds a bridge between the
life sciences and the physical sciences.
Best,
Bill P.
According to PCT, perceptual
functions are neural networks that convert sensory input into neural
signals.
PCT has very little to say about
this. It’s only a proposition in the same book. Where can I find a more
detailed rendition.
How is this perceptual function created and what gives one the condition to say
this is a blueprint of any kind. In the same book mention is made of Pribrham
and his holographic brain model which I know a bit about. But this still doesn’t answer the question.
This reference signal is beginning to feel a bit arbitrary to me. Ofcourse it
fits the model of CS unit.
My image of a perceptual function that perceives honesty, for
example, takes sensory input, such as the visual image of a sales
person
giving a sales pitch, and converts it into a neural signal, the
scalar value of which is the perception of the level of honesty of the
pitch. Let’s say that the output of this perceptual function can
range
from
0 to 100 impulses/sec, where 0 is a perception of dishonesty and
100 is a perception of perfect honesty. Then one can set a reference
“blueprint” for high honesty by setting a reference signal to, say,
This is a huge jump now, in
inferences about the brain, its function and things like honesty (your choice
of value), fairness, cooperation, freedom, love and all other so called values.
Else where in PCT I have read that the reference signal is a goal or an
intention (on this I may be under correction because I am unable to find it
now). This also then slides back to this elusive blueprint now a neural
function.
Ofcourse goals and intentions must come from somewhere, they can’t just
be there in the blueprint and the neural functions (whatever this may be).
If the reference signal can also be an intention then we have a whole new
dialogue.
Theories of mind (mostly input/ouput models as per your definition) try capture
this so
called blueprint, but you seem to be accepting it as a given. That’s the
whole point of psychological theories to explain the mind (blueprint).
When you say healthy, in terms of what reference are you using?
In terms of the person’s own hierarchy of references for their own
perceptions. A healthy person is (from my point of view) one who is
managing to keep all controlled perception under control, maintaining
a low ambient level of error in the entire control hierarchy. My
spreadsheet hierarchy illustrates this; it comes “out of the box” as
a
healthy hierarchy, keeping all perceptions at all three levels close
to their constant (at level three) or varying reference values.
Are there healthy people and unhealthy people in terms of PCT?
No, but you can use PCT to give those terms some coherence. I
think of
mentally
healthy people as people who have a very low level of ambient
error in their nervous systems; that is, they are able to keep
all the
perceptions
they want to control under control. Since the PCT theory
of emotion suggests that chronic error results in physiological
changes that are experienced as things like anxiety, depression or
anger, it seems like a person who is not keeping their perceptions
under control (and, thus, experiencing high levels of error) is a
person who feels like they themselves have “problems”. These are the
people who are the most likely to seek professional help. Indeed, one
of my best friends, who was always kind of tentative about PCT, when
and became a counselor and was surprised to find that nearly
everyone
who
came to him for help said that they felt that their life was "out
of control".
Sure that’s hardly a proof though. “
out of control” People seek help mainly to reduce mental anguish
and I ‘m sure that not all feel out of control. But that’s besides
the point. Lets forget about these other question I want to get to the root of
the reference signal.
[From Bill Powers (2008.12.28.1127 MST)]
Rick Marken (2008.12.28.0930) –
… perceptual functions are neural networks that convert sensory input into
neural signals. My image of a perceptual function that perceives honesty, for
example, takes sensory input, such as the visual image of a sales
person giving a sales pitch, and converts it into a neural signal, the
scalar value of which is the perception of the level of honesty of the
pitch. Let’s say that the output of this perceptual function can range
from 0 to 100 impulses/sec, where 0 is a perception of dishonesty and
100 is a perception of perfect honesty. Then one can set a reference
“blueprint” for high honesty by setting a reference signal to, say,
90.
Good explanation but it needs a little more detail. A basic principle used in
the PCT model is that all perceptions are one-dimensional. They can only have
one scalar value at a time, so can be expressed as a number. Every perceptual
input function, therefore, receives multiple input signals and produces just
one perceptual signal as its output.
An alternative model would say that a perceptual input function receives multiple
inputs and produces multiple outputs representing a multidimensional
perception. That seems to fit experience better – when we perceive something
like a “chocolate soda” this is not just a “how much”
perception, but very much a “what kind” perception with all sorts of
qualities.
After puzzling over these two possibilities for a long time, back in the 1950s,
I saw what the answer had to be. The key to the problem lies in awareness, and
its ability to register more than one perceptual signal and more than one level
at a time. The alternate model above seems better because it includes many
attributes of the chocolate soda: its name, the chocolate flavor, the
fizziness, the straw sticking out of the standardized soda glass, and so on.
What finally made up my mind was realizing that each of these attributes is a
perceptual signal! Awareness receives information not just from one perceptual
input function but from many, and not from just one level but many. The above
descriptions are about conscious experience. Awareness is mobile and its
scope varies; it can include more perceptual signals or fewer, more levels or
fewer. The field of consciousness is the intersection of awareness with a set
of perceptual signals in various places in the hierarchy.
So now I could go back to the first model, a much simpler model in which each
perceptual signal represented just one dimension of experience at one level,
and say that conscious experiences included the outputs of many of these
simpler perceptual input functions. The actual workings of the hierarchical
model, however, did not involve multidimensional signals, but only simple
frequency-coded signals in which the frequency indicates the degree to which
the perceptual input function is recognizing the one attribute to which it
responds. Later on, I found that this was the same organization that Oliver
Selfridge had assumed in his “pandemonium” model: the demon that
yelled the loudest won the identification contest. If I show you a mouse, your
elephant-perceiving perceptual input function responds a little because there
are four legs, a nose, a tail, a gray color, and movement – but the elephant
recognizer responds a whole lot more.
There’s some potential confusion or interaction here between the ideas of
awareness encompassing multiple input signals, and higher-order perceptual
input functions also encompassing – receiving – multiple input signals. An
elephant-perceiving input function would receive signals representing how much
noseness there is, how much sizeness, how much tuskness, and so on, and respond
the most when these input signals had the right proportions. Then the
higher-level input function would generate a signal indicating that a lot of
elephantness is present. So how is that different from awareness experiencing
all the signals representing size, nose, color, and so on and seeing the
elephant that way?
The difference is exactly in how many details there are and at what levels they
exist. When you remember seeing elephants at a circus in your childhood, you
may just remember, as we say, that you saw elephants. The memory carries
a sense of elephantness but without any details: what size, how many, how big,
headed which way, silent or noisy, fragrant or smelly. The single elephantness
impression is the recording of the higher-order perceptual signals being
replayed into the higher-order perceptual signal channel. But if you saw the
elephants half an hour ago, it’s likely that a lot of details (no one of which
is an elephant) come to mind, including color, sound, smell, motion, shape,
relationship, events – all the lower-level signals that are classified at
level 6 (I propose) of the hierarchy and named “elephant” – the name
being a configuration perception included in the same category.
In short, both the higher-order perception of elephantness and the lower-level
perceptions of attributes are received by awareness and make up the whole
experience of a real, present elephant. If the higher-level elephant signal is
not present but the lower-level attribute signals are present, we see a pattern
but we don’t “recognize” it. Maybe some elements are missing or faint
or in peculiar relationship. It’s like looking at that pattern of black and
white blobs for a while, seeing them perfectly clearly, but not seeing the
Dalmatian dog. When imagination finally supplies the critical missing elements,
the Dalmatian recognizer finally wakes up and says “Oh, that’s mine. Here,
look, look, look!” And suddenly it’s a whole dog with spots.
Combining awareness with a one-dimensional model of perception thus gives us
the best of many worlds. The automatic functioning of the control processes is
easiest to explain at the neural level where all perceptual signals are
one-dimensional, but the combining of the signals into higher-level, but still
one-dimensional, signals explains how conscious experience fits in. Of course
that leaves us with a new mystery, the mystery of what awareness is, but I
think it’s a net gain.
All this came together in the 1950s and early 60s. Yet for some reason I held
back on the ideas that much later became the method of levels, in which
awareness plays a central role. All right, I just didn’t see the connection,
though it’s perfectly obvious now. It was actually Tim Carey who gave me that
last sense of reality that lets me talk more confidently about these things
now. He insisted that the PCT model was absolutely essential to understanding
the method of levels, and of course I agreed since that was good for the ego.
But now I see: it’s all part of the same model, though one big piece still
looks rather ghostly.
Best,
Bill P.
No virus found in this incoming message.
Checked by AVG - http://www.avg.com
Version: 8.0.176 / Virus Database: 270.10.1/1867 - Release Date: 12/28/2008
2:23 PM
