# Perceiving error

**URL:** <http://discourse.iapct.org/t/perceiving-error/8700>\
**Category:** CSG1993\
**Created:** [September 22, 1993, 1:44pm UTC](http://discourse.iapct.org/t/perceiving-error/8700 "1993-09-22T13:44:28Z")\
**Posts on this page:** 8\
**Page:** 1

<div class="post-metadata">

**Author:** ![William\_T\_Powers2](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/w/c67d28/32.png) [@William\_T\_Powers2](http://discourse.iapct.org/u/William_T_Powers2)\
**Post date:** [September 22, 1993, 1:44pm UTC](http://discourse.iapct.org/t/perceiving-error/8700/1 "1993-09-22T13:44:28Z")

</div>

[From Bill Powers (930922.0710 MDT)]

Greg Williams (930922)

Nice to hear from you again.

I didn't get much chance at the meeting to ponder the materials  
you brought. The exerpts you cite now raise some questions in my  
mind.

> 1. "When driving an automobile, the HUMAN OPERATOR perceives a  
> discrepancy or error between the desired state of the vehicle  
> and its actual state. The car may have deviated from the center  
> of the lane or may be pointing in a direction away from the  
> road. The driver wishes to reduce this error."

This says explicitly that what the driver perceives is an error,  
implying that the error exists outside the driver. If the error  
exists outside the driver, then the comparison process must also  
take place outside the driver, interpreting certain relationships  
of the car to the road as errors and presenting this interpreted  
result to the driver's senses.

In PCT, what is perceived is simply the relationship of car to  
road that exists at any time. It is neither correct nor  
incorrect; it is what it is. The perception of this relationship,  
inside the driver, is compared with a reference signal, also  
inside the driver. The driver can select any state of the  
perceived relationship as a reference condition, and manipulate  
the car to achieve the reference relationship -- for example, car  
straddling the lane-separator line, car crosswise in the road,  
car leaving the road altogether (as when turning a corner), car  
crashing into a bridge abutment in a suicide attempt. What comes  
in to the driver's senses is not an error signal, but simply a  
report on the current state of affairs as the perceptual  
equipment represents it. There are no errors or "discrepancies"  
represented in the perception.

> 2. (regarding tracking of time-varying inputs in general:) "It  
> is assumed that the human 'intends' to produce a given  
> position."

The implication here is that the intended position is the actual  
position of the car relative to the actual road (etc.), outside  
the driver. Putting "intends" in quotes suggests that the author  
doesn't treat intention as a real phenomenon, but just as a way  
of speaking. I don't see that Wickens has made it clear that an  
intended position is an intended \_perception\_ of position, as  
opposed to an actual position objectively defined. Nor is it  
clear that Wickens identifies an intended position as an  
adjustable reference signal inside the controlling system.

Wickens doesn't seem to grasp the situation as we do in PCT.  
Milsum, I agree, did better.

Perhaps what we need is a survey of all the major manual control  
experts' way of describing simple control situations, to get an  
idea of how many understand that only perceptions can be  
controlled, and that intentions and error signals exist inside  
the driver. From your citations from Wickens, I'd say he may  
understand control theory but he flunks PCT, although perhaps he  
redeems himself in other parts of the book.

We'll put in an interlibrary request for the Latash book.

> **···**
>
> ---------------------------------------------------------------  
> Best,
> 
> Bill P.

---

<div class="post-metadata">

**Author:** ![Bruce\_Abbott](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/ba8739/32.png) [@Bruce\_Abbott](http://discourse.iapct.org/u/Bruce_Abbott)\
**Post date:** [September 27, 1996, 1:42am UTC](http://discourse.iapct.org/t/perceiving-error/8700/2 "1996-09-27T01:42:36Z")

</div>

[From Bruce Abbott (960926.2040 EST)]

> Rick Marken (960926.1530) --

> > Bruce Abbott (960926.0820 EST)
> 
> > Under this condition, to state that "the experimenter is \_not\_ a 'source of  
> > information as to error' " (emphasis mine) is, strictly speaking, incorrect.
> 
> > > Bill Powers (960926.1430 MDT) --
> 
> > > We should all kick ourselves. KR is effectively part of the controller's  
> > > input function
> 
> Nice observation. But KR is still not "information as to error". At best, it  
> is what you say: "experimenter-supplied information about the state of an  
> otherwise unobservable controlled variable." Saying that it is "information  
> as to error" suggests that error is something we perceive;

I think we're getting into a language problem here. If the participant has  
adopted the experimenter's reference (or if the experimenter has adopted the  
participant's reference), and the experimenter is reporting the deviation of  
the CV from reference, and the participant takes corrective action on the  
basis of that deviation, isn't that the same as if the participant observed  
the CV, compared it to the reference, and took corrective action on the  
basis of that deviation? If so, then it would not be incorrect to say that  
the experimenter is providing "information as to error" (deviation of CV  
from reference)? Perhaps I can perceive neither the reference position nor  
my current position, so I delegate to the experimenter the job of comparing  
the two and reporting the error to me. I then do what I always do with such  
errors when computing them myself.

And who says we can't perceive error? Are you asserting that when I wish to  
be at A, and perceive myself to be at B, that I can't perceive the error in  
my position? And act on it?

> As Buddha might have said: we do not perceive error: error is a result of  
> comparing what we perceive to what we \_want\_ to perceive.

No, wise Buddha would have perceived the error in stating that we do not  
perceive error, and thus would have said no such thing!

Regards,

Bruce

---

<div class="post-metadata">

**Author:** ![Bruce\_Gregory9](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/43a26b/32.png) [@Bruce\_Gregory9](http://discourse.iapct.org/u/Bruce_Gregory9)\
**Post date:** [September 27, 1996, 2:36pm UTC](http://discourse.iapct.org/t/perceiving-error/8700/3 "1996-09-27T14:36:05Z")

</div>

[From Bruce Gregory (960927.0935 EDT)]

(Bruce Abbott 960926.2040 EST)

> \>Rick Marken (960926.1530) --
> 
> \>\>Bruce Abbott (960926.0820 EST)  
> \>  
> \>\>Under this condition, to state that "the experimenter is \_not\_ a 'source of  
> \>\>information as to error' " (emphasis mine) is, strictly speaking, incorrect.  
> \>  
> \>\>\>Bill Powers (960926.1430 MDT) --  
> \>  
> \>\>\>We should all kick ourselves. KR is effectively part of the controller's  
> \>\>\>input function  
> \>  
> \>Nice observation. But KR is still not "information as to error". At best, it  
> \>is what you say: "experimenter-supplied information about the state of an  
> \>otherwise unobservable controlled variable." Saying that it is "information  
> \>as to error" suggests that error is something we perceive;
> 
> I think we're getting into a language problem here. If the participant has  
> adopted the experimenter's reference (or if the experimenter has adopted the  
> participant's reference), and the experimenter is reporting the deviation of  
> the CV from reference, and the participant takes corrective action on the  
> basis of that deviation, isn't that the same as if the participant observed  
> the CV, compared it to the reference, and took corrective action on the  
> basis of that deviation? If so, then it would not be incorrect to say that  
> the experimenter is providing "information as to error" (deviation of CV  
> from reference)? Perhaps I can perceive neither the reference position nor  
> my current position, so I delegate to the experimenter the job of comparing  
> the two and reporting the error to me. I then do what I always do with such  
> errors when computing them myself.

When we delegate to someone else the responsibility for  
determining the error, do we not tacitly shift our goal? I  
think this is what happens whenever a teacher "steers" a  
discussion or asks questions to which he or she knows the  
answers. The students' goals then become, not to understand in  
terms of their own base of knowledge, but to say what they  
believe the teacher wants to hear. Students are very good at  
mindreading. In many ways they understand how to test for  
whatever the teacher is controlling for much better than the  
teacher ever understand how to test for what each of the  
students is controlling for. Perhaps it is because they work  
cooperatively and the teacher is alone with a different set of  
goals.

Bruce

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [September 27, 1996, 2:11pm UTC](http://discourse.iapct.org/t/perceiving-error/8700/4 "1996-09-27T14:11:00Z")

</div>

[From Bill Powers (960927.0645 MDT)]

Bruce Abbott (960926.2040 EST)--

> I think we're getting into a language problem here. If the participant has  
> adopted the experimenter's reference (or if the experimenter has adopted the  
> participant's reference), and the experimenter is reporting the deviation of  
> the CV from reference, and the participant takes corrective action on the  
> basis of that deviation, isn't that the same as if the participant observed  
> the CV, compared it to the reference, and took corrective action on the  
> basis of that deviation?

A long time ago, and for reasons that are partly neurological and partly  
arbitrary, I set the rule for myself that perception is associated ONLY with  
afferent pathways in the brain. By that I mean the pathways that begin with  
sensory neurons and build higher and higher levels of representation on the  
signals from those neurons. A strict interpretation of that rule says that  
we do NOT experience reference signals or error signals, or the signals  
emitted by any level to lower levels (efferent signals).

It was the strict application of this rule (which dates back close to the  
beginning in the period with Bob Clark, 1953 - 1960) that led to proposing  
the imagination connection. If only afferent or upgoing signals can be  
experienced as perceptions, how do we know the values of reference signals  
before we put them into effect? We obviously do: I can tell you what my aim  
is before I carry it out. But that means we are perceiving reference  
signals, and the rule says that we do not perceive reference signals, which  
are efferent signals.

It took a while before the answer popped up. In order not to violate the  
basic rule, somehow we have to get the reference signal information into the  
afferent channels so it can be perceived. The simple-minded way to do this  
(my normal way) was to route a copy of the reference signal through a  
short-circuit into the perceptual input function of the system that is  
providing the reference signal. Then the input function of the higher system  
receives a signal just like the one it would receive if the lower system had  
actually made its own perceptual signal match the reference signal it is  
receiving from above. Normally the higher-level input function receives  
copies of lower-level perceptual signals; substituting the reference signals  
sent to each contributing lower-level system would create the same situation  
that would hold if the lower systems were all controlling perfectly.

This immediately suggested "imagination," because imagination has the same  
sort of properties that this connection would have. First, the  
short-circuited reference signal would enter the same input function that  
the lower-level perception would normally enter, so the signal would receive  
the same interpretation it receives in the normal mode of operation. Second,  
the conversion of the reference signal into a perception would be immediate  
and effortless -- control of the perceptions would be perfect. Third, it  
would be possible to imagine perceptions in this way which the lower-level  
systems could not actually provide, either because doing so would be  
physically impossible (imagine lifting your car over you head) or because  
you don't happen to be in the required environment (while you're driving to  
work, imagine having a conversation with your boss). Fourth, the terms in  
which we imagine are exactly the same terms in which we perceive. Such  
properties certainly fit what we call imagination. Incidentally, brain  
research has recently shown that animals which are anticipating the  
appearance of a stimulus show activity in the same perceptual channels that  
are involved when the stimulus actually appears, which is strong support for  
this picture.

It was only a short step to get memory into this. If efferent signals from  
one level of systems are translated by a memory function into the actual  
signals received as references by lower systems (a feature which would take  
care of translating from the terms of a higher system into the terms of a  
lower one), then the same connection could explain remembering as well as  
imagining. The memory aspect of reference signals comes in when you "do the  
same thing again." If the "thing" is turning left at the third traffic  
light, what you are repeating is obviously the perception of turning left,  
not the actions that lead to this result in the traffic of today, and with  
the timing of today's arrival at the traffic light which could be either  
green or red (or broken) when you get there. What you're repeating is the  
\_perception\_ of turning left, not the \_actions\_, which means that the  
perceptions must be remembered somehow and used as reference signals. But if  
you can do that, you can also simply remember without acting, because the  
source of the reference signals is memory and the imagination connection can  
route the output of memory into the perceptual input functions that will  
give that output the correct interpretation.

So we get imagination, remembering, and "doing the same thing again" all  
from the basic postulate that all perception is of afferent signals. Quite a  
harvest from planting a single seed. And of course we also get planning and  
dreaming -- model-based control is the "modern" term.

OK, I wanted you to see the structure of the model that explains why I want  
to keep the postulate that all perception is associated with afferent  
channels only.

> And who says we can't perceive error? Are you asserting that when I wish to  
> be at A, and perceive myself to be at B, that I can't perceive the error in  
> my position? And act on it?

This is how levels of perception get into the act. I'm on the left side of  
the street (A) and I wish to be on the right side of the street (B). How do  
I perceive being on the right side of the street when I am not there? By the  
above reasoning, I must imagine being there. Imagining provides a signal  
(actually a lot of signals, because we're really talking about multiple  
systems operating in parallel) that is like what I would experience if I  
were in position B, on the other side of the street. Normal-mode perception  
provides similiar signals representing being in position A, the position  
where I am. I can now \_perceive the difference in positions\_ at a higher  
level of perception. One of the positions is real, the other imagined. And I  
can perceive the relationship between them.

The relationship is a certain distance between A and B. And what is the  
\_reference\_ relationship? ZERO distance between A and B. So the higher  
control system turns this error (which I do not perceive) into an action  
that moves me from A to B and corrects the relationship error.

If the control-system error signals were the errors we perceive, then we  
would always have to reduce all errors to zero. It's very easy to see the  
target in a pursuit tracking task as setting the reference position and the  
cursor position as being the controlled variable. Obviously, good tracking  
requires minimizing the "error" in cursor position. If you've become  
satisfied with this interpretation, it comes as a great shock to see the  
controller suddenly start keeping the cursor two inches to the right of the  
moving target. I always try to administer this shock when demonstrating  
control, as with the rubber band experiment. The point is that the  
controlled variable is a spatial \_relationship\_, and cursor-on-target is  
just one of a possible range of relationships that could be maintained. It's  
not an error in the control-system sense, it's just a distance. If you  
always pick the reference condition of zero distance you have a degenerate  
case, because zero error does happen to coincide with zero distance between  
target and cursor. But if you pick a different distance like two inches to  
the right for the cursor, zero error corresponds to a NON-ZERO distance  
between cursor and target. And then you can see that while the spatial  
relationship is perceivable, the error signal is not.

Best,

Bill P.

> **···**
>
> ---------------------------------------------------------------------------

---

<div class="post-metadata">

**Author:** ![Richard\_Marken](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/r/9fc348/32.png) [@Richard\_Marken](http://discourse.iapct.org/u/Richard_Marken)\
**Post date:** [September 27, 1996, 4:10pm UTC](http://discourse.iapct.org/t/perceiving-error/8700/5 "1996-09-27T16:10:14Z")

</div>

[From Rick Marken (960927.0900)]

Me:

> Saying that it [KR] is "information as to error" suggests that error is  
> something we perceive;

Bruce Abbott (960926.2040 EST) --

> I think we're getting into a language problem here.

I think it's quite a bit deeper than that. If the PCT model is correct, then  
saying that any perception (including KR) is "information as to error" is  
not a language problem; it is a gross misrepresentation of how a control  
model actually works. I think it also represents a failure to understand the  
deepest insight PCT gives us about the nature of human nature, viz. that  
perceptual experiences have no inherent value; they are just representations  
of external reality. Different people will seek or avoid the same perception  
because people \_want\_ different perceptions. The value (good, bad, error,  
correct) of perception is determined by what people want (references for  
percpetion); not by the perception itself.

> If the participant has adopted the experimenter's reference (or if the  
> experimenter has adopted the participant's reference), and the experimenter  
> is reporting the deviation of the CV from reference, and the participant  
> takes corrective action on the basis of that deviation, isn't that the same  
> as if the participant observed the CV, compared it to the reference, and  
> took corrective action on the basis of that deviation?

In this case the KR does, indeed, correspond to the "error" in the  
participant/controller. It's the same as the situation that exists in a  
tracking task where the subject has agreed to control the distance between  
target and cursor, keeping that distance at zero. The distance between  
target and cursor corresponds fairly closely to the actual error signal in  
the subject's brain. You might be content call the distance between target  
and cursor -- the KR in the tracking task -- "information as to error". But,  
as Bill Powers (960927.0645 MDT) noted "if you've become satisfied with this  
interpretation, it comes as a great shock to see the controller suddenly  
start keeping the cursor two inches to the right of the moving target".  
Suddenly the subject seems to be ignoring a rather consistent piece of  
information about error.

You might say that calling the KR "information as to error" is still  
legitimate; after all, you understand PCT so you know that when the cursor is  
being kept two inches to the right of the moving target the subject's  
reference for the distance between target and cursor has changed. So the  
"information as to error" that exists in the perception has changed. Now  
there is information as to error only when the cursor is \_not\_ two inches to  
the right of the target.

But I don't like this linguistic slight of hand because it obfuscates a  
beautiful and powerful PCT observation with language. The beautiful and  
powerful observation is that the "value" of an experience (like KR) is  
determined by the subject; value does not "inhere" in the experience itself.  
The tracking experiment shows that the "value" of a particular experience  
(the distance between target and cursor) can change; at one time zero  
distance between target and cursor is "valued"; at another time a two inch  
distance between target and cursor is "valued". This change in value occurs  
(according to PCT) not because of any change in the experience itself but  
because of a change in the subject; a change in one of the subject's goals  
(reference signals). This change in goal changes what was once a valued  
perception (zero distance between cursor and target) in one that is actively  
avoided.

Conventional psychology has tried to attribute changes in the "value" of  
experience to the external environment: it has done this by giving  
environmental events imaginary properties: valences, affordances,  
disciminativeness, reinforceingness, informativeness, etc. It is very hard  
for PCT to make headway when people "take it for granted" that environmetal  
events actually have many of these properties. When you say that KR  
provides "information as to error" you are playing into these misconceptions;  
you allow people to cling to the belief that something in the world (KR) has  
properties like "informativeness"; something other than the subjects  
themselves that determines the "value" of an experience (like the experience  
of hearing the words "correct" or "incorrect").

> No, wise Buddha would have perceived the error in stating that we do not  
> perceive error, and thus would have said no such thing!

I think the error is yours -- and I think it is a profound one. (But,  
characteristically, it is quite consistent with your desire to see no  
conflict between PCT and conventional psychology). It is a profound error  
not only in terms of understanding PCT but in terms of real life dealings  
with people. If you actually believe that KR is "information as to error"  
then you must be quite puzzled and annoyed when you give KR and people don't  
treat it as "information as to error". You have given me a ton of  
"information as to error" (such as your statement above about the wise  
Buddha) and, yet, I have done nothing to correct my ways. Maybe there just  
isn't enough information as to error in your KR? Or could it be that we  
have different references for the same perception (PCT) - - a perception  
that, in itself, is not right or wrong, good or bad, correct or in error?  
Naaa. That would mean that individual people (rather than an external  
environment or god) determine the value of what they experience. We certainly  
can't have THAT now, can we;-)

Best

Rick

---

<div class="post-metadata">

**Author:** ![Bruce\_Abbott](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/ba8739/32.png) [@Bruce\_Abbott](http://discourse.iapct.org/u/Bruce_Abbott)\
**Post date:** [September 28, 1996, 11:10pm UTC](http://discourse.iapct.org/t/perceiving-error/8700/6 "1996-09-28T23:10:59Z")

</div>

[From Bruce Abbott (960928.1810 EST)]

> > Bruce Abbott (960926.2040 EST)

> > I think we're getting into a language problem here. If the participant has  
> > adopted the experimenter's reference (or if the experimenter has adopted the  
> > participant's reference), and the experimenter is reporting the deviation of  
> > the CV from reference, and the participant takes corrective action on the  
> > basis of that deviation, isn't that the same as if the participant observed  
> > the CV, compared it to the reference, and took corrective action on the  
> > basis of that deviation?

> Bill Powers (960927.0645 MDT) --

> A long time ago, and for reasons that are partly neurological and partly  
> arbitrary, I set the rule for myself that perception is associated ONLY with  
> afferent pathways in the brain. By that I mean the pathways that begin with  
> sensory neurons and build higher and higher levels of representation on the  
> signals from those neurons. A strict interpretation of that rule says that  
> we do NOT experience reference signals or error signals, or the signals  
> emitted by any level to lower levels (efferent signals).

Thanks for the explanation, Bill (only partly quoted here). I understand  
your reasons for adopting this assumption. But I do have my own perspective  
to add.

I think we have to be very careful here not to equate perception with  
"experience." Experience implies a conscious perceiver, yet many perceptual  
signals apparently are never consciously perceived. Perception itself  
results from the analysis of neural input, and typically that input begins  
at sensory receptors, although it may arise from other pathways, as in your  
suggestion that perception can arise from the activation of stored patterns  
called memories. The nervous system is organized such that sensory signals  
are (with a few minor exceptions) conducted along afferent pathways from the  
sensory receptors toward those spinal and/or brain structures that perform  
the analyses; at higher levels in the brain the patterned activity in these  
structures may give rise in some unknown way to conscious experience.

Limiting the term "perceptual signal" to cover only those signals conducted  
through afferent pathways is probably as good a solution as any when  
breaking down the system for purposes of analysis. A perceptual signal  
arising from a muscle spindle follows an afferent pathway to the spinal  
cord; it remains a perceptual signal under this definition whether or not it  
is conducted beyond the immediate, low-level control system of which it is a  
part. But so, it seems to me, would an afferent signal arising from the  
comparator, whose magnitude represented the difference between reference and  
perceptual input signals (other wise known as "error"). Whether such  
perceptual signals exist in the nervous system and participate as inputs to  
other levels of the system is, it seems to me, an empirical matter. It is  
an assumption of PCT as opposed to a deduction from it. Furthermore, it  
matters not a whit whether such signals ultimately become part of  
"experience," as the failure to be experienced is no sure criterion for  
rejecting a given signal as perceptual.

The entity labeled "error" in a control system is nothing more than a  
difference (discrepancy) between two signals; discrepancies can be  
perceptual signals. (I would propose, however, that the term "error signal"  
be reserved for describing the function of this particular discrepancy  
within the control loop, which is to drive action.) What is an error signal  
\_within\_ a control system would only be a another perceptual signal running  
up the afferent pathways, so far as higher-level systems are concerned. If  
such an afferent signal existed and if that signal improved the functioning  
of some system further upstream in the brain, then I don't see any reason  
why over the course of evolution such a connection could not have been  
preserved. Whether this arrangement exists or not is a matter to be  
disclosed by future research; their nonexistence is a PCT assumption, not an  
established fact, even though that assumption may contribute a certain  
elegance to the PCT formulation.

With respect to memory and imagination depending on the same perceptual  
centers that analyze sensory input of the same qualitative type as the  
remembered or imagined "image," this view is actually inherent (at least to  
my mind) in the "doctrine of specific nerve energies" proposed around 100  
years ago by Mueller, which states that the quality of a sensation depends  
on the area of the brain in which the neural impulses are analyzed and not  
upon anything inherent in the impulses themselves; e.g., neural signals  
arriving at the auditory cortex produce sounds because they arrive at the  
auditory cortex; rerouted to the visual cortex, the same signals would  
produce sensations of light. It would be wasteful of brain tissue if the  
internally-generated patterns we interpret as dreams or imaginings had to be  
produced via different analytic structures than those used for primary  
perception. And the evidence that we do use these same structures for both  
types of perceptual experience is, I think, overwhelming. For example,  
destruction of the cortical region responsible for color analysis eliminates  
a person's ability, not only to see in color, but also to dream in color,  
and even to imagine what colors "look like."

> > No, wise Buddha would have perceived the error in stating that we do not  
> > perceive error, and thus would have said no such thing!

> Rick Marken (960927.0900) --

> I think the error is yours -- and I think it is a profound one. (But,  
> characteristically, it is quite consistent with your desire to see no  
> conflict between PCT and conventional psychology).

Rick, your ability to read other people's minds is not nearly as good as you  
believe it to be. In the two years I've been subscribed to CSGNET, I've  
never once indicated that I have a desire to see "no conflict between PCT  
and conventional psychology." But thanks for demonstrating how the Test for  
the controlled variable can fail miserably.

Bill Powers tells us that an inability to perceive error is simply a PCT  
postulate -- an assumption to be tested rather than a known fact. As it is  
quite evident that I (although evidently not you) can perceive discrepancies  
or differences between perceptual signals, the question as to why PCT  
assumes that "error" cannot be perceived is certainly one worth asking.

> It is a profound error  
> not only in terms of understanding PCT but in terms of real life dealings  
> with people. If you actually believe that KR is "information as to error"  
> then you must be quite puzzled and annoyed when you give KR and people don't  
> treat it as "information as to error".

What I actually said is that \_if\_ the experimenter and participant adopt the  
same reference value, then what the experimenter perceives (and reports as  
error) is the same as what the participant perceives (and acts upon) as error.

In your response, you ignore the qualifier and respond as if I had stated  
that KR "is" information as to error, at all times and under every  
circumstance. This is not my position, and the rest of your attack strikes  
not at me but at, of all things, that ubiquitous CSGNET straw man -- the  
only person here actually holding that position.

Regards,

Bruce

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [September 29, 1996, 5:29am UTC](http://discourse.iapct.org/t/perceiving-error/8700/7 "1996-09-29T05:29:59Z")

</div>

[From Bill Powers (960928.2036 MDT)]

Bruce Abbott (960928.1810 EST)--

> I think we have to be very careful here not to equate perception with  
> "experience." Experience implies a conscious perceiver, yet many perceptual  
> signals apparently are never consciously perceived.

This is theoretically true, and I have said as much. I make a distinction  
between perception (a signal in an afferent pathway) and awareness  
(experiencing that signal). The fact that lower-level systems continue to  
control when you are aware only of higher-level processes shows that  
perceptual signals must exist outside awareness.

There seem to be perceptual signals of which one is not aware, but of which  
one can become aware by a shift of attention. These have some empirical  
basis, in that it seems reasonable that the signals were there all the time  
(someone speaking to you while you're reading something fascinating), but  
they suddenly appear in conscious experience when attention shifts to them.

Perceptual signals of which nobody is ever aware, however, create a dilemma.  
Who knows about them? If any process in the brain is \_permanently\_  
inaccessible to awareness, it must be purely hypothetical; there's no way to  
verify that such signals could be classed as "perceptions." All we know  
about anything exists in the realm of consciousness, at least while we are  
attending to it. What is \_never\_ consciously experienced isn't part of what  
we consciously know. Which simply means we don't know about it.

> What is an error signal \_within\_ a control system would only be a another  
> perceptual signal running up the afferent pathways, so far as higher-level  
> systems are concerned.

Yes, if you want to propose a model organized like that you're certainly  
free to do so. However, you have to make it work. Have you ever tried to  
simulate a two-level system in which the information reaching the  
higher-level input function is the error signal in the lower system? I have,  
and it didn't work as a controller. The problem is that the error signal is  
then part of the basis for adjusting the reference signal, which directly  
affects the error signal; you get a closed loop that doesn't include the  
rest of the lower-level world. What happens is that the reference signal is  
adjusted until the error signal matches the higher-level reference, and  
what's happening in the lower-level world doesn't get controlled. No matter  
what the lower-level perception is, the reference signal gets adjusted until  
the error is some particular amount. So either no action results, or the  
action is maintained constant no matter what happens to the perception. That  
might be an interesting feature for accomplishing something or other, but it  
doesn't accomplish control of perception.

We're really talking about an empirical question here. Do we, in fact,  
experience error signals? In the cases we're been talking about, confusion  
is caused because the controlled variable is a relationship between two  
lower perceptions, so we can verbally label that relationship an "error" if  
we want to. This gets even more confusing when we're relying on someone  
else's description of what THAT person considers to be an error -- that is,  
whatever relationship between two perceptions that person is calling an error.

Let's consider a simpler case. Suppose you simply want to close your fist.  
Just before you close it, you can feel and see the configuration of the hand  
that exists. But where is the error, when you decide you want it closed?  
This is very different from the case where you can see a target just as well  
as you can see a cursor, and where you can clearly see the distance between  
them. Those are obvious present-time perceptions that seem perfectly clear  
and real -- which go away when you close your eyes.

But what are you perceiving when you decide you want your hand closed? All  
you actually perceive with your senses is the hand, open. You don't have a  
visual image of a closed hand that is also arising from your senses, so you  
can see the relationship between the closed and open hands. The only true  
perception is of an open hand. You can, with a slight effort, \_imagine\_ your  
hand closed, but that is nothing like perceiving a closed hand with your  
senses (at least it's not for me). But you don't need to do that in order to  
close your hand: all you need to do is "will" it closed, and it closes. When  
I do that, I get no hint of either the reference signal or the error signal.  
I know they must be there, theoretically, but I don't experience them.

It's just as clear that you don't experience efferent signals. The signals  
that make your muscles get tense are, as far as I can tell, totally  
inaccessible to awareness. All you ever experience is the consequence of the  
muscles contracting. You have to deliberately imagine the consequence in  
order to create anything like experiencing an efferent signal -- and by my  
hypothesis, you're then routing the reference signals into the perceptual  
functions, so you're still being aware of afferent signals only.

I agree that what is called an error signal in a control system is just a  
difference between two signals. But here the two signals are not both  
perceptual signals originating at lower levels, and the organization of the  
system is such that the difference is always brought as close to zero as  
possible. When the difference is between two signals from lower systems, it  
is no longer an "error" except in comparison with some reference signal that  
specifies how large the difference should be.

Of course that reference level could be zero, leading to all our  
difficulties. But it doesn't HAVE to be zero; it can have any value. In the  
game of warmer and colder, you as the behaving system can change your mind  
and start trying to get the experimenter to keep saying "colder." No matter  
what the so-called "error" is, you can redefine the target value of the  
error to be any nonzero amount if you want. So this "error" is really just a  
controlled perception like any other, and the REAL error signal is still not  
being perceived.

Best,

Bill P.

---

<div class="post-metadata">

**Author:** ![Hortideas\_Publishin1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/h/f6c823/32.png) [@Hortideas\_Publishin1](http://discourse.iapct.org/u/Hortideas_Publishin1)\
**Post date:** [December 10, 1992, 11:32am UTC](http://discourse.iapct.org/t/perceiving-error/8700/8 "1992-12-10T11:32:00Z")

</div>

> From Greg Williams (921210)

> Mark Olson (921208)

> Basically, Greg Williams sent me a post which clarified  
> things for me really well--its alot like what I've been saying all along,  
> (but didn't quite know it--funny how that works). Whether it's perceptions  
> or error that is controlled is dependent on what LEVEL from which one is  
> speacking. From an individual ECS level, its percpetions,no doubt--the  
> math shows that conclusively. But from the perspecive of teh whole  
> organism (the organism as an organism, as I said earlier) its error.

> Basically both are equally correct from each perspective. Before I thought  
> both were correct but one was more correct.

> Rick Marken (921209.0900)

> So I ask you (and Greg) to explain how "perspective"  
> (or "level") influences the results of the test for controlled variables. If  
> the "error" variable can be "not controlled" from one perspective but  
> "controlled" from another, then I could be coming to the wrong conclusion  
> when I say "this variable is not controlled" because I happen to have done  
> the test from the "wrong" perspective. It also would be possible to conclude  
> that, say, the difference between curson and target is "controlled" from  
> one perspective and "not controlled" from another. If this is possible, then  
> methodologists should know how to tell what perspective they are  
> doing the test from and how it influences their results.

That will teach me to post privately. Below is what I sent to Mark last  
Monday:

> From Greg Williams (921207 - direct)

Hi Mark -

I think you're being conned by the math.

The basic PCT loop doesn't control error, but the function of such loops  
is "ultimately" to keep intrinsic error from getting out of hand. The loops  
control perceptions (not error), seen individually, but the purpose of the  
SYSTEM of individual loops, seen from a higher perspective, is to control  
error (which is what surviving in order to reproduce is all about). This is  
the same argument-of-perspectives which I've brought up several times: the  
current-time perspective (emphasized by Bill P.) vs. the historical  
perspective emphasized by Skinner. The loops-as-currently-structured control  
perceptions AND are themselves modified when necessary to control errors.  
"Ultimates" are too verbal for my taste, yet I can appreciate the "ultimate"  
nature of error control in "shaping" the control structure, vs. the  
"pragmatic" nature ("proximal" nature?) of the moment-to-moment workings of  
the loops in the structure, controlling perceptions (if they're able).

So, I think the problem is that Rick leaves out where the basic loop came

Hope this makes at least a little sense,

[end of post to Mark]

I see the function of the entire hierarchy as doing whatever it is able at a  
particular time to keep intrinsic errors within a tolerable range (with the  
result that the organism keeps living). If, at some time, that becomes  
impossible, then the hierarchy changes (reorganization). So at the "systemic"  
level, intrinsic errors are being perceived and if any of them get too large  
for too long, then actions are taken until they get back within "spec" (or,  
failing this, the organism dies) -- that's negative feedback control of the  
perception "too much (of this or that intrinsic) error." Still, each loop in  
the hierarchy doesn't control a perception of error. So error is controlled by  
the reorganization process, but not by the reorganized loops. Those are the  
different levels for which The Test shows that different "kinds" of variables  
are being controlled (actually, in the former case, error is ALSO a perceptual  
variable; it isn't in the latter case).

Rick, do you think reorganization does NOT have to perceive error? If not, how  
does it know when to start and stop?

Best wishes,

Greg

P.S. Ooops, I see I slightly goofed above -- should have said "that's negative  
feedback control of 'tolerable (this or that intrinsic) error.'" The reference  
signal is for acceptably low error.

> **···**
>
> from: from the need to control (intrinsic) errors.