[From Rupert Young (2016.12.10 21.30)]
Here's my long-winded response (I must get a life!), of some
thoughts on my perspective.
(Martin Taylor 2016.12.06.13.53]
2: [RY] [MT]... more than once you have said that you
KNOW that there are “no dots below the line”, … [RY] It’s not a
matter of knowing in the sense of having knowledge of the external
world, but that it is a conclusion which follows from my
understanding of PCT; particularly of a non-representational view
of controlled variables.
[MT] For me, this suggests that your understanding of PCT differs
profoundly from mine, so I suppose we should investigate why. I
start with what I take to be the minimalist canonical perceptual
control loop and ask some questions about it.
Questions (they are leading questions, but they may have
unexpected answers).
A: Is this a fair representation of a control loop?
B: Is the set of three functions "Perceptual Function",
“Comparator”, and “Output Function” the internal part of the
control loop (and the grey area the external part of the control
loop). (For convenience, I call this set of connected functions an
“Elementry Control Unit” or “ECU”.)
C: Are the values labelled "p", "r", "e", "o" and "d" the values
of variables with conventional names respectively “perception”,
“reference”, “error”, “output”, and “disturbance”. Properly
speaking, I should use notations parallel to p(t) to make clear
that these are variables as functions of time, but I hope that
would be unnecessary clutter, and we can just take it for granted
– though I find taking things for granted has not always panned
out in this thread:-)
I'll go along with these first three. But, of course, we are talking
about the model rather than a real control system, in which there
are no labels as there are here. There are just neural signals, and
all that matters to the real system is that particular signals are
brought to values which help it survive. We impose labels upon the
arrangement in order for us to make some sense of the functional
operation of the nervous system and for us to try and understand
what is going on. However, once we do that there is a danger that
the words we use may have connotations for us that do not completely
accurately portray the actual situation, and may even mislead us to
conceptualise the system in a way that is erroneous. This may seem
obvious, but is relevant to later points. I’ll also just mention
that when I am talking about perception I just mean neural signals
and not the conscious experience that may go with them; that, I
think, is a distraction to this discussion.
D: Is the value labelled “s” or “q.i” a variable?
This is where I disagree. It is not (as a rule) a single variable,
in my view, but a set of variables, which are inputs to the input
function, as shown by Rick the other day with his diagram with the
set of “v” variables. I believe the Big Powowski himself described
q.i as a convenient fiction; something that helps us understand
control systems, but is not something (a single variable) that
actually exists in the external world. (NB: I’ll use “external
world” rather than “environment” as I mean external to the whole
perceptual hierarchy of the organism, as I know you (Martin)
sometimes use “environment” to mean external to a particular control
system, but still within the hierarchy.) There may be, I suppose, a
single variable at the very lowest level, but even there do
intensities correspond do single raw external variables or
combinations thereof?
My answers to these questions are all "Yes", but I'm
guessing yours are not, because if they are, my version of logic
suggests that since the value of “p” is strictly determined by the
value of “s” or “q.i”, therefore “perception” is a direct
representation of whatever name you want to give to the variable
labelled “s” or “q.i”.
I see this is where the problem, the difference in our perspectives,
lies, with the concept of “representation”. I think this may be
derived from the connotations associated with the use of the label
“perception”, which may be thought of as denoting something in the
environment.
But what is representation and what do we mean by it? One way the
term might be used is give a name to something, such as saying this
neural signal “represents” fear. But that is just self-referential
labelling and not what we are talking about here.
What we mean here is that an internal (neural) signal corresponds to
(“represents”) an external variable; “representation” is a property
of perceptions. But I think this is erroneous and not the right way
to think about the concept. It is not some physical law or a
property of perceptual systems, but merely a perspective that helps
us understand control systems. Sometimes! But sometimes not, and can
actually be downright misleading. Sometimes perceptions can be
thought of as representing variables in the external world, but
other times there are no corresponding external variables. One way
of testing this (mentally) is by removing the perceiving system and
thinking about whether there is still an external variable that
corresponds to the perception.
Here's some examples to illustrate how I think about this:
-
When taking a shower we turn the dial to control the feeling
of comfort related to the temperature of the water. I don’t
particularly have problem with saying that the perception
represents the water temperature. But only as a loose
description of there being some sort of correlation between the
feeling and the temperature. Temperature is actually, I believe,
an indication of the speed of molecules within the water, in
this case. Now I don’t know how the perception is constructed,
but it doesn’t seem likely that it is a direct representation of
the molecule speed. It seems more likely that it is constructed
from a set of sensations and intensities of reactions on the
skin. The perception itself is a variable on the right-hand side
of the perceptual function combining a set of inputs, but in
itself does not exist independently of the perceiving system.
Incidentally, in this case, if the showerer were to leave the
bathroom and leave the shower running there would still be a
variable (water temperature) that exists, independently of the
subject (or any) perceiving system.
-
When we are speaking face-to-face with someone we control the
proximity at a comfortable level. You could say that the
proximity perception represents the distance between the two
people, but this is just a way, in this case, of interpreting
what they are doing that is useful to an external observer. And
an external observer would be able to take measurements related
to the perception. If the subjects were to leave there would no
longer be a variable in existence; it is dependent upon the
subject perceiving system.
-
When viewing the Ames room the subject perceives a rectangular
room, but in this case there is no rectangular room. In other
words, there is no external variable that the perception
represents. But there is the set of inputs that give rise to the
perception. The perceptual function creates the perception here
irrespective of whether or not there is correspondence in the
external world. In this case there isn’t even anything (a single
variable) in the external world that could be labelled as the
represented variable.
The first two examples give the false impression that perceptions
directly represent equivalent complex variables in the external
world rather than being constructions of sets of raw external
variables. Representation is a perspective that may be of use (a
convenient fiction) in some circumstances, but its use is limited
and is not a general rule when it comes to the signals we call
perceptions.
Here are some other examples that are closer to the third example
above:
-
Pain is a variable that only exists as a perception, internal
to the perceiving system. It seems ludicrous to me to talk of a
variable “pain” that exists in the external world.
-
My robot example mentioned previously perceives, and controls,
a smoothed sum of sensor values. There is no such variable
external to the system. It is entirely dependent upon the
internal hierarchical structure.
-
Likewise with perceptions such as honesty, fear, anxiety,
hunger, religious beliefs, love etc. There are no corresponding
variables in the external world. They are internal constructions
for the purpose of being controlled to aid survival, not
properties of the external world.
-
With random dot stereograms there is no "dolphin" variable in
the external world. The only time a “dolphin” variable exists is
when the focus of the eyes is adjusted to form the perception,
and the only place that it exists is internal to the system, on
the right-hand side of the perceptual function. Though what does
exist externally are the constituents that become inputs to the
function, by way of the sensory interface.
-
Something simple like a plankton could be seen to be
controlling variables that represent the direction of light and
the intensity of light, but ultimately it is controlling its own
energy levels (which it gets from light) which is not an
external variable. It may well help us observers to make sense
of the behaviour of the plankton to think of the former two
variables as “representative” of external variables, but to the
organism it is irrelevant. They are just lower level variables
by which it can control its higher level goal.
-
We often hear about the way women dress (short skirts, not
wearing a veil) as being responsible for being perceived as
“provocative”. I think this situation falls under the
representation misconception, that somehow there is something in
the external world that is being “detected” rather than a
subjective perception constructed according to the particular,
prejudicial make-up of a particular perceiver.
The signals being controlled, at anything above the very lowest
level, derive from functions of sets of lower level values, and only
exist as unique constructions that do not correspond to external
variables. Some, though, may have the appearance of representing
external variables, to an observer; but this is an illusion, a
convenient fiction. In my opinion, it is misconceived to apply the
representation perspective as a general rule or property of the
signals we call controlled variables.
So, to summarise a few points:
-
Some perceptions may appear to correspond to external
variables, others do not.
-
I suggest we forget, for the moment anyway, about
representation, and even about signals as being described as
perceptions.
-
Rather, let's just think of them as inputs (controlled
variables), without any imposed meaning. That is, outputs of
what we have been calling “perceptual” functions which are
inputs to higher systems, but also under control themselves.
-
These functions are not about extracting, recognising or
detecting variables from the external world, but about
constructing variables to be controlled.
-
Organisms control these inputs in order to survive.
-
To this end the main thing under control is energy levels.
-
The purpose of all other controlled variables, is to help with
achieving that main goal.
-
Input Control Theory works perfectly well without thinking
about these inputs as being “representative”.
This may seem something of a philosophical discussion, but I think
it crucially relevant to both how we understand control systems and
how we go about constructing artificial control systems. If we think
all controlled variables must represent some variable in the
external world then we are going to neglect from our implementations
variables that do not, such as my sensor sum example. As the latter,
in my view, correspond to the majority of controlled variables then
our ability to build advanced artificial control systems is going to
be severely restricted, if not impossible.
This all seems sort of obvious to me, may be it does to others. On
the other hand, if one has generally thought of perceptions as being
representative of variables in the external world then this may seem
alien (or crazy) to you and perhaps your first inclination is to
disagree. But hopefully I’ve done enough so that you understand my
perspective even if you do not agree. However, I request that you
don’t dismiss this out of hand, but mull it over and give some time
to think about. Hopefully, after some consideration, you may also
see that it makes sense, and provides a more liberal and less
restricted understanding of the theory than might otherwise be the
case.
Rupert
