Misunderstandings and distortions. See DISPUTE.PCT
Unedited posts from archives of CSG-L (see INTROCSG.NET):
This thread deals with myths, misunderstandings, and distortions offeedback, cybernetics and PCT. The term Devil’s Bibliography was suggested and stuck. Thus this file is called DEVIL’S.BIB.
See file TOUGH_SE.LL for the following post by Tom Bourbon as a suitablestarting point:
Date: Thu Oct 22, 1992 6:21 pm PST
Subject: PCT popularity; Why 99%?
Date: Sat Oct 24, 1992 6:19 pm PST
Subject: Objections to PCT
[from Gary Cziko 921025.0200 GMT]
To Bill Powers, Greg Williams & Tom Bourbon and other interestedparties:
When I try to discuss PCT with “mainstream” psychologists, two objections often come up: (a) feedback is too slow for many behaviors; (b) deafferentated animals can behave with no sensory input.
I know that both of these subjects have been discussed on CSGnet in the past, but I wonder if there exists published or unpublished papers which address these issues with more rigor. If these don’t exist, perhaps they should.
In the meantime, I remember Bill having made comments about the “speed”objection (and perhaps the deafferentation objection too) and Tom having made comments about the deafferentation studies and I would appreciate if they could summarize their arguments. From Greg I would expect some arguments that the lowest levels may not be fast enough for feedback control with the control effectuated by higher levels sending output commands and perceiving the resultsof the commands (although I don’t want to start Bill vs. Greg on yet anothertopic before they’ve settled the current one).
If these objections against PCT are so common, then perhaps PCTers shouldhave objections ready against the objections. –Gary
Date: Sun Oct 25, 1992 7:31 am PST
Subject: Objections to PCT
[From Bill Powers (921025.0800)] Gary Cziko (921025.0145) —
> When I try to discuss PCT with “mainstream” psychologists, twoobjections often come up: (a) feedback is too slow for many behaviors; (b)deafferentated animals can behave with no sensory input.
The first objection is a myth tracing back to cybernetics; the second is astraw man argument.
If you think of feedback as something that follows after the end of abehavior, then of course feedback is too slow. When you realize that feedbackactually starts at the instant that action begins and continues throughout theaction, however, that lag disappears.
Another “slowness” commonly cited is reaction time. The reaction timepeople usually think of is 200 milliseconds, the reaction time of a saccade orof a motor response to a sudden visual stimulus. This is far longer than actualdelays in kinesthetic control systems. The delay in the lowest spinal-cordcontrol loops is 9 or 10 milliseconds, and in brainstem loops only about 50milliseconds.
What laymen don’t realize is that even WITH lags, a control system can bestabilized by the use of proper temporal filtering in the loop. William Ashby,who did much to start these myths, was a psychiatrist; he didn’t know anythingabout stabilizing control systems. In fact, a properly stabilized controlsystem with a lag can reach equilibrium in one reaction-time,given good filtering. See my Psych Rev article “spadework” [QuantitativeAnalysis of Purposive Systems: Some spadework at the Foundations of ScientificPsychology, reprinted in Living Control Systems, Vol I] where I lay out therequirements for achieving this by using a “slowing factor.”
Another related myth is that a stimulus-responsesystem must be faster than a control system. A properly-filteredcontrol system is always at least as fast as, and usually much faster than, astraight-throughSR system using the same components but without feedback. The reason is nothard to understand.
If you want a response proportional to a stimulus without feedback, youhave to adjust the gain so that in the steady state, the output is of therequired magnitude in relation to the input. The device responds by producingan output that rises exponentially to a final value given a step input (allreal devices have at least this kind of lag). The maximum possible speed ofresponse without feedback is thus set by the inherent slowness of the physicaldevice.
If we use negative feedback from the output to the input, we can arrangethe feedback ratio so that in the final steady state, when the feedback cancelsthe input, the output is again of the required magnitude. Now, however, we cangreatly increase the amplification in the device itself, which is in theforward part of the loop. This does not speed the device up; it still takes aslong as before to reach, say, 90% of the final output value. All it does ismake that final output value much larger.
Thus when a step input occurs, the initial response of the device is suchas to approach an output value that is many times as great as the desiredamount of response. The output begins to rise very much faster than in the casewithout feedback. If there were no feedback, the result would be an enormousovershoot of the desired output value. But as the output increases, so does thenegative feedback. When the final state is reached, the negative feedback iscanceling most of the input, and the output becomes exactly the desired amountwith no overshoot at all. But the time taken to reach the final state is only afraction of what it would be without the feedback.
This is exactly what H. S. Black discovered in 1929. He found that vacuumtube amplifiers with a certain inherent gain and bandwidth could be used toachieve not only far more stable gain but a much wider bandwidth, through theuse of negative feedback. This knowledge never got into cybernetics, and thusnever got into psychology, and thus failed to inform the mythmakers in thesefields. Control engineers didn’t read the psychological literature, so theynever set the record straight.
The fact is that a design with negative feedback is almost always faster inresponse than a design without negative feedback.
This objection is a straw man. There has been a great deal of grisly andclumsy research aimed at disproving a claim that nobody ever made: that withoutfeedback, there can be no behavior. Taub and Bizzi and others, being ignorantof control theory, misinterpreted what control theorists have to say aboutfeedback, and set out to disprove their own misinterpretation. A simple phonecall to the right people could have saved decades of effort and a whole lot ofmisery.
Consider the following control diagram.
| ref sig
<–ext feedback <–action
WITH feedback the perception is made to match the reference signal. As aresult, when the reference signal changes the output changes so as to make theperception change as required for a match. We see the output as”behavior.”
Now deafferent the diagram: remove the perceptual signal:
| ref sig
<–ext feedback <–action
Will there be behavior when the reference signal changes? Of course therewill be; there is still an intact path from the reference signal, through thecomparator, to the effector. Loss of the negative feedback will resultinitially in greatly exaggerated outputs and wild instability; this has beenobserved by everyone who has studied the effects of lesions and injuries onafferent paths. Loss of feedback doesn’t give you NO behavior. It gives youMORE behavior. Control theory actually predicts exactly the kind of thing thatis seen when the feedback path is interrupted.
There are, however, other feedback paths to higher centers; visual,tactile, and so on. These are not removed by deafferentation. Also,deafferentation is commonly done by cutting the dorsal roots of the spinalcord; this leaves the “auxiliary” pathways in the ventral roots intact.
In any event, deafferented animals are given a post-operativerecovery period of, if I remember right, about 16 days before they are tested.During this time, the higher systems learn to control their perceptions usingthe above un-fed-backoutput system, and lower their own loop gain so that the reference signalsgoing to the deafferented system have a smaller range of change. Thiseliminates the gross instability that resulted from the loss of kinestheticfeedback and provides some semblance of normal behavior. What we see isbehavior that is basically open-loopwith respect to kinesthetic control, but closed-loopwith respect to visual control or tactile control. The kinesthetic controlsystem can no longer resist mechanical disturbances, beyond the amount ofresistance created by the elasticity of muscles. A load deflects the limb andthe deflection is not corrected. Dynamic stability is poor (in fact, in oneexperiment by Bizzi that I saw, the animal’s forearm was strapped to a pivotedboard which had frictional contact with a table underneath it: I suspect thatthis was required in order to keep the arm from oscillating and overshooting.The authors did not explain why this was necessary in order to demonstrate”unchanged behavior”).
The whole deafferentation fiasco was motivated not by a desire tounderstand how behavior works, but in order to defend the conventional viewagainst the threat posed by control theory, as the researchers understood thatthreat. If these researchers had bothered to study control theory first, to seehow it would explain the behaviors they were studying, they would have realizedthat control theory fits the observations very well indeed, whereas to makeconventional theory fit them a great deal of cheating is needed.
Best, Bill P.
Date: Tue Dec 15, 1992 3:05 pm PST
Subject: Where are the Goals
[FROM: Dennis Delprato (921215)]
Only skimmed over reviewer’s comments on Rick’s most recent clash withtradition. Two quick points:
1. On goals: One impediment to people like the reviewers grasping PCT isthe tendency to treat PCT as saying something about what they know as goals(fine). BUT the problem is that they do not get away from the conventionalnotion that the goals are “out there.” I suggest that this one importantfactor in many not grasping what the likes of Rick are saying.
2. On feedback control: The crude reactions to feedback control especiallyfrustrate me. Mainstreamers seem to take one or more of the followingpositions:
a. Feedback control is not important in psychological behavior.
b. It is important but we know all there is to know about what’s going onhere. Let’s get on with the important stuff.
c. Feedback? Oh, you mean reinforcement. This has been beaten to death. Or–Iam quite up on reinforcement, am continuing the work of the great learningtheorists.
d. Feedback control –I agree it is important for motor skills, but we are not interested in thisarea.
e. Feedback? –Too mechanical, OK for machines but not for….
f. Feedback? That’s information on how well one is approximating a goalthat is out there in one’s external environment.
g. Feedback? Control? I don’t know what your are talking about and thatain’t all ’cause I don’t give a damn either.
This is not a very sophisticated classification of reactions to feedbackcontrol, but it does begin to give some idea of how it is that K. U. Smith’sand now Powers’s work tends to be met with wide yawns.
Date: Wed Dec 16, 1992 5:42 pm PST
Subject: Point of View
[FROM: Dennis Delprato (921215)]
Rick Marken’s recent replies to reviewers of his “Blindman” paper broughthome what I suggest is another major roadblock to comprehension of PCT. Recallhis major thesis:
“If organisms are in a negative feedback situation with respect to theenvironment, then their behavior will APPEAR to be SR, reinforcement andcognitive when it is actually NOT –it is CONTROL OF PERCEPTION”. (921213)
This is obvious to CSG-Lparticipants, but a most remarkable position to just about everyone else who isinvolved in bio-behavioralscience. What makes it so difficult is that it rules out the classical scienceidea of the independent observer. The above fundamental of PCT is based on aparticular “point of view.” The point of view is NOT one of the conventionalindependent observer/experimenter of the classic independent variable-dependentvariable framework of mainstream psychological science. In fact, basically,Rick’s point is that as long as researchers stick to the classic methodology,they will never detect the very different picture from the one traditionyields. From its point of view, behavior is best described as S–>R,motor program –>movement, ….
The issue seems to be somewhat related to what physics went through inmoving from classic mechanics/physics to what the professional physicsliterature refers to as the new (or modern, or relativistic, or quantum)physics. As physicists more and more went into the microscopic world, theyfound, to their dismay, that the physical events could not be described andknown independently of the behavior of the observer, including the particularsof the conditions of observation (e.g., “measuring instruments”). They didn’trealize it then, and still don’t, but they discovered that the dichotomybetween a physical world and a psychological world (so much a part of ourcultural tradition) no longer made sense, except as fiction. Notions such asindeterminacy and complementarity evolved to help thinkers cope with theconfusing conclusion that a physical world independent of the observer, andthus an observer independent of the physical world as well, no longer fit intoformal scientific formulations. Only the vernacular, but not the formalscientific language, allowed preservation of the traditional dualisms.
As Dewey and Bentley (in Knowing and the Known, 1949) put it, physicalscience was forced to adopt transactional procedures of inquiry in which”transaction” refers to the “full ongoing process in a field where all aspectsand phases of the field, including inquirers themselves, are in commonprocess.”
Although it is possible to find hints of recognition of the need fortransactional procedures of inquiry in psychology, as we full well know,classic independent variable -dependent variable with independent observer is the only approach to inquiryeven taught at the highest levels of psychological training. (And it is mostunfortunate when one has to resort to “naturalistic observation” or to”correlational” methods. But at least one can work like the dickens to keep theobserver *independent* in these cases.) Rick is asking quite a bit of hisreaders. I imagine few even recognize what they are being asked to consider.If we examined them in great depth, I’ll bet that the most astute will get asfar as something like, “This is your (Dr. Marken) point of view. I suppose youbelieve it sincerely and with good reason, but I Cannot buy it. I cannot getinto your shoes/head/mind.”
I fear that wider acceptance of PCT views will require more directconsideration of the inevitable role of the observer in psychological inquiry.When this has been addressed in the mainstream literature, the observer hasbeen taken as a creator of data and knowledge, with the result of preservingmaterial-spiritual dualism. I believe the PCT alternative is one in which knowledge isrelative to the observer. My suggested emphasis may seem unnecessary, but theequations do not interpret themselves, as should be obvious by now.
Date: Sun Dec 20, 1992 9:08 am PST
Subject: Misstatements & Other Basics
[FROM: Dennis Delprato (921220)] >(Bill Powers (921218.1500)
>RE: Feedback is too delayed.
> Dennis, would you be willing to become a repository for citations fromthe literature containing misstatements about feedback control, PCT,etc.?
Pleased to, especially given that I seem to have already begun this out ofmy own curiosity.
Note another major roadblock that you bring up in the following: “I am aprofessional control-systemengineer / I have a close friend who is a real control-systemengineer / I have a high IQ and studied control system engineering / … andyou simply are not getting it right at all. Too bad, too, since the idea ofapplying control system theory to humans is an excellent way of showing howpsychology is nothing but physics.”
>Bruce Nevin (921218.1324) —
> That Latin saying developed into a most interesting and relevantdiscussion. It is surely true that our most profound problems in introducingPCT come from those who think they already have a grasp of what feedback andcontrol are about. You’ll remember that a year or so ago we had a participanton the net who was a “real control-systemengineer.” He obviously understood control systems –but he absolutely could not accept the statement that control systems controltheir inputs! He eventually bade us farewell, saying in a perfectly friendlyway that he just couldn’t go along with this strange way of looking at controlsystems, but good luck to us.
Date: Mon Dec 21, 1992 11:49 am PST
Subject: Re: Martin to Rick on Shannon
From: Tom Bourbon (921221 10:15) [Martin Taylor 921218 18:30]
Make a simple offer …! Edited… Dag 930606
Tom to Martin (in the present):
Neither the model of the control system or the environmental phenomena withwhich it interacts need be linear. Bill has published and posted onintroducing nonlinearity into the PCT model and into the environment. So hasRick. I haven’t, but I have tested the effects of nonlinearities in thecoordinated systems: The models continued to function at the same level ofrealism. I will try to put together a post on that topic, in the style of mypost a few days ago on adding disturbances to various signals in the controlsystem.
In the meanwhile, I wonder why so many people continue to assert that PCTmodels are necessarily linear and cannot explain and predict events when thereare nonlinearities in the system or the environment. Where do these ideas comefrom? Why won’t they go away? (Dennis Delprato: If you are starting acollection of false assumptions and assertions about PCT, this certainly isone. We should compare collections –mine goes back a few years.) Everyone who clings to that assumption should readBill’s “spadework” paper in Psych. Review (1978 –14 years ago folks) [in LCS I] where he discussed various blunders in thehistory of cybernetics. That is also where he quantitatively demonstrated theease with which a PCT model maintains control in the presence ofnonlinearities.)
Until later, Tom Bourbon
Date: Mon Dec 21, 1992 8:44 pm PST
Subject: Prediction as Feedforward
[from Gary Cziko 921222.0430 GMT] Dennis Delprato and Bill Powers:
I suppose here’s another misunderstanding concerning control systems thatwe might want to add to our library.
In discussion the application of control system models to understanding”motor control” with a physiological psychologist, he was arguing that theability to predict meant that feedforward was taking place. The example he usedwas predicting the movement that an object would take and using that knowledgefor tracking (like how it’s easier to point to a the end of a swinging pendulumthan to a fly caught in the shower stall with you).
I don’t think I did a very good job at trying to explain how trackingpatterns can be seen as controlling a higher-levelperceptual variable. Perhaps someone can help could help me out with this. Itis also related, I believe, to the discussion between Taylor and Powersconcerning what a control systems has to be able to “predict” in order tomaintain good control.
Date: Tue Jan 05, 1993 5:17 pm PST
Subject: Devil’s Advocate
[From Bill Powers (930105.1530)] Greg Williams (920105) —
> For the tracker to “respond” to the “discriminative stimuli,” all thatis necessary is for him/her to be able to see the cursor movement, NOT to”tell… WHAT THE DISTURBANCE IS.” If the cursor is seen to be moving away fromthe target position –due to the net COMBINATION of handle position and net disturbance, of course –then the tracker responds by moving the handle in the direction (determinedpreviously in practice, via “reinforced” learning) which moves the cursor inthe direction toward the target position.
How about “the tracker sees an error between the cursor position and itsintended position, and responds by moving the handle at a velocity proportionalto the amount of the difference and a direction corresponding to the directionof the difference?” This is a verbal description of the organization of thecontrol system.
Your way of putting this assumes that the intended position of the cursorrelative to the target is AT the target. It is perfectly possible to move thecursor so it remains a fixed distance to either side of the target. This makesthe definition of a discriminative stimulus somewhat difficult, because at thatspecified distance from the target, most of the time, one can see –nothing. The stimulus now has to be defined as the distance between the cursorand an arbitrarily-locatedempty place in space, or alternatively as the distance of the target from thatempty place minus the distance of the cursor from that empty place. No matterhow you put it, the discriminative stimulus has to be imaginary.
This mistake has been made many times in the past –the view assumes that some “salient” (meaning obvious-to-me)aspect of the situation is the reference condition, forgetting that thiscondition is just one point on a scale, and therefore not realizing thatcontrol could take place relative to any position on that scale. This is howpeople have concluded that reference signals come from the environment. That’sanother myth that got launched in the ’50s.
The cursor position and velocity always reflect the ongoing behavior of thedisturbance PLUS the ongoing behavior of the handle. If the cursor beginsmoving slowly to the right, this could indicate that the disturbance hasstarted pushing it to the right a little faster than the handle is pushing itto the left, or that the handle has started pushing it to the left a littleslower than the disturbance is pushing it to the right. The informationrequired to make even this qualitative judgment is not contained in the cursorposition or velocity. You must perceive your own handle movements directly andestimate how the cursor would be moving and where it would be positioned ifyour handle were the only influence.
In Demo1 there is a phase in which the difference between compensatory andcontrol behavior is illustrated. In compensatory behavior the “cursor” on thescreen shows the disturbance magnitude, not the actual cursor position. Thetask is to estimate where to put the handle at each instant so that the effecton the now-invisiblecursor would keep it from being disturbed. This is impossible on the face ofit, so the demonstration shows a trace of what happened to the real cursorduring the run (afterward), and you can also alternate with controlling thereal cursor so you can pay attention to how your hand moves and learn how muchit needs to move and where the center of movement is. By using all this (higher-levelfeedback) information over may trials, you can actually improve yourperformance in the compensatory phase quite a lot. You can, with a lot ofpractice, get the RMS error in the invisible cursor position down to only about10 times what it is when you can’t see the disturbance but can see thecursor.
Even with all this practice, you can’t estimate your handle’s effect on thecursor well enough to achieve the kind of control you get without having to payattention to the handle at all and without any direct information about thedisturbance magnitude. In the compensatory case you don’t have to estimate theamount of disturbance by comparing felt handle position with seen position ofthe cursor. You are given an exact quantitative picture of the disturbancemagnitude. And you still can’t achieve the performance of a control systemwithin less than a factor of 10 worse.
Furthermore, when the disturbance is shown on the screen as a pointer ATTHE SAME TIME YOU ARE CONTROLLING A VISIBLE CURSOR, your tracking performanceis not measurably different in most cases when the disturbance information iseliminated (this is not in Demo1 but I have done the experiment). The only casein which some measurable difference can be seen is when the participant paysattention to the disturbance information and tries to use it to improvecontrol. In that case, the quality of control deteriorates sharply until theperson ceases to pay attention to the disturbance information.
All these demonstrations, which I have actually done and which are easilyreproducible, show that the person is not making any use of information aboutthe disturbance, either directly when it is available on the screen, orindirectly by estimation of expected handle effects on the cursor.
These facts about control are easily demonstrated, but to understand themyou have to think quantitatively. As long as you are allowed to talk about”cursor movements” and “handle movements” and “target movements” you can glossover these quantitative facts because the language doesn’t specify HOW MUCHmovement there is in relation to other movements. In a subtle way, you’re usingthe known outcome of the experiment to provide just the meaning for the generalterms that is needed to make them fit the observations. It would not bepossible to go the other way: if you didn’t know how the experiment came out,describing the relationships in terms of “movements” and “positions” and othersuch qualitative notions wouldn’t tell you anything about which way the cursorwould move. The cursor movement is a small difference between two variablesthat are changing over a range of perhaps 20 times as much (RMS comparisons).Just saying that the disturbance moves the cursor to the right while the handletends to move it to the left leaves the direction of the actual cursor movementundefined.
I tell you: the disturbance is changing to push the cursor to the leftwhile the handle is moving to push it to the right. Which way, pray tell, doyou predict that the cursor will be moving?
In your Devils’s Advocacy, you have brought out exactly what is wrong withconventional objections to PCT. It is the qualitative nature of the argumentsthat makes it seem that alternative explanations fit the facts. Qualitativedescriptions are crude enough to allow for any outcome at all –or its opposite. Qualitative explanations, if cleverly enough constructed, areunfalsifiable. This is the attraction of generalization and qualitativedescription: you can’t be wrong.
> One last time: don’t take high correlations as THE sign of stimulus-responserelationships.
Why not? That’s what THEY do, isn’t it? This brings out the main reason wecan’t talk to conventional behavioral scientists. If you show them a trackingexperiment, the first thing they will do is look for high correlations: thisbehavior is a response to that stimulus. When we carefully set up theexperiment to show what actually happens to the correlations, what do they do?Do they say “Oh, migosh, it looks as though I have the wrongexplanation!”?
In a pig’s eye. They immediately back off, and say that this situation ismore complex than it appeared, and requires a different explanation. Theyabandon the simple analysis of simple experimental data and start talking aboutvague effects of discriminative stimuli and reinforcements, all of whichsomehow have exactly the effects required to account for the experiment –for the cursor going up a little here, down a little there, and wiggling justso in between while the handle traces out an almost perfect mirror image of theinvisible disturbance.
The real problem is that such people don’t have any idea of what a realexplanation amounts to. They have given up on science.
Best, Bill P.
Date: Tue Jan 12, 1993 12:12 am PST
Subject: feedback too slow
Reading around in the Osherson, ed. `foundations of cognitive science'(1989) I came across the claim that feedback is too slow to solve inversekinematic & dynamic problems for fast movements. Where can I read aboutwhy this claim is false or irrelevant (e.g., true only for certain kinds ofhighly skilled movements that people practice enough to make it plausible thatthey have elaborate feedforward schemes for).
Date: Tue Jan 12, 1993 2:48 am PST
Subject: slow feedback reference bungled
[Avery Andrews 9201112.2108]
The reference to the excessive slowness of feedback is actually:
Jordan and Rosenbaum (1989) `Action’, in Posner, ed. _Foundations ofCognitive Science_, pas. 731, 746. This paper also cites some very suspect-lookingwork by Kelso and others that is supposed to support a concept of `coordinativestructures’ (what seems to be going on is low level configuration/relationshipcontrol between articulators).
I guess the obvious question to ask about any movements that do look toofast for feedback is how they go under novel or unpredictable dynamicconditions. E.g. how does the concert pianist make out with lead weightsattached to her fingers.
Date: Mon Jan 11, 1993 1:10 pm PST
Subject: Who’s got the generative model?
[FROM: Dennis Delprato (930111)] Greg (930109) Bill (930108.0800)
I believe the history of science shows that worthwhile new ideas advancewhen they are contrasted to previous ones that workers find of value. For thisreason, I am enthusiastic about Greg Williams’s recent attempts to push somebasic operant theoretical accounts to the limit vis-a-visPCT.
>> But he [Skinner] was fixated on environmental control of behavior,and was forced to conclude that behavior is controlled by its consequences,even though the only CLEAR relationship he could see was that of consequencesbeing controlled by behavior. I have always considered this to be his mostintellectually dishonest ploy.
> You have to remember that he construed current control of consequencesby behavior (which he freely admitted) as itself having “come under thecontrol” of the organisms’ history –of consequences in the past.
And through which mechanisms does this history operate? The operanttheorist is forced to stay descriptive. Fear of mysterious nonspatiotemporalinner processes leads them to posit no underlying process. Admirable perhapswhen all that was available was mysterious. Wide open territory forPCT.
>> From another standpoint, the behaviorist COULDN’T characterize theexperimental setup correctly. To do so would be to see that the stimulus is notan independent variable. The assumption is that the stimulus varies, and as aconsequence of that the response varies. To measure the response, onearbitrarily varies the stimulus, so the stimulus has a known value or patternthat is independent of the behavior. If the stimulus is defined so it dependson the response, it’s impossible to perform this manipulation (without breakingany actual feedback loop that’s present).
> I think you are putting words into the mouths of at least somebehaviorists (including Skinner), but if you can produce some documentation tosupport your claim about what they say, I’m ready to be corrected. A whileback, I posted some quotes from Skinner which contradict the notion that hethought a stimulus could not be affected by the responding organism. Note thateven in a feedback situation, the stimulus can be manipulated to a degree,because control isn’t perfect.
The trouble may be that they stop with a description of the experimentalset up (i.e., procedure). They take the procedure (e.g., S dee-Response-Reinforcer,coupled with deprivation or the like) sufficient for explanation.Incorporation of a control system account requires that one go beyond theobvious details of the procedure, including history of reinforcement and verbalstatements that “feedback is involved.”
>> Skinner saw the reinforcer as a consequence of behavior. But beingunable to give up the idea that the environment controls, he then treated thisconsequence as an independent variable, and said that it controls the behavior.To be sure it controls only FUTURE behavior, but with his blind spot he neversaw the obvious implication: that the BEHAVIOR which produces this consequencecontrols (“controls” meaning influences) the future behavior via the apparatus.To see this loop whole would have meant giving up the concept that theenvironment determines behavior, and that, above all, he was unwilling todo.
> Seeing the whole loop, with BOTH environmental and organismicinfluences on output, is EXACTLY the middle way between environmentalism andorganismism which I was arguing for some time back.
Bill is right. Skinner always came back to the environment as the ultimateindependent variable. Although he and many followers have verbally stated thatenvironment determines response and response determines environment (stimulus),in practice, this has amounted to nothing. True, some followers (e.g., Baum)have gone a bit farther than did Skinner, but Bill can tell us aboutthese.
>> No, the behaviorist will yell “Foul!” when I point out that the so-calledstimulus is not an independent variable.
> That’s what I’m not so sure about. I wish we had a real behaviorist onthe net. Maybe Dennis could speak to this issue?
Some behavior analysts are getting away from stimulus as true independentvariable and response as true dependent variable. BUT, as with above comment,they do not go anywhere with this thinking. Close but so far.
In my opinion, the major stumbling block for behavior analysts as far asPCT is the requirement to take S-Rand R-Sas SIMULTANEOUS. This is very difficult for two basic reasons: (1) our cultureteaches lineal thinking and (2) the damn procedure (itself a product of linealbiases) is lineal (ess D then R then reinforcer). The only way to get S-Rand R-Ssimultaneous is via theory, then this poses the added difficulty of changingthe status of the response consequence.
Hope I’ve said something to help Greg. My perspective suggests there is agreat opportunity for one to prepare an interesting paper entitled somethinglike “From Feedback Functions to Perceptual Control Systems.”
Date: Tue Jan 12, 1993 9:13 am PST
Subject: slow feedback
[From Rick Marken (930112.0800)] Avery.Andrews (930111.1905) —
> I came across the claim that feedback is to slow to solve inversekinematic & dynamic problems for fast movements. Where can I read aboutwhy this claim is false or irrelevant
I think Bill Powers posted a wonderful exposition on the “feedback is tooslow” nonsense. I don’t have it available but maybe someone could re-postit; it was really excellent. The answer to the above claim is simply “how doyou know that people solve inverse kinematics to make movements of ANY kind –slow or fast; maybe they just control their perception”?
> This paper also cites some very suspect-lookingwork by Kelso and others that is supposed to support a concept of `coordinativestructures’
Welcome to the wonderful world of high powered motor control nonsense. Twopapers in “Mind readings” are attempts to show, experimentally, that the leapinto ‘coordinative structure’ models may have been a bit precipitous.
> I guess the obvious question to ask about any movements that do looktoo fast for feedback is how they go under novel or unpredictable dynamicconditions. E.g. how does the concert pianist make out with lead weightsattached to her fingers.
You’ve got it! The question has been asked, and answered, by PCT researchand models, many times. But no one pays attention because they already KNOWthat PCT can’t work (because feedback is too slow). If you want to read moreamusing statements about feedback being “too slow” made by authoritativeleaders in the study of human movement control, try the article by Abbs andWinstein in M. Jeannerod (ED) Attention and Performance XIII, Hilldale,Erlbaum, 1990 (I just pulled this off the shelf and found the reference bylooking in the index under “feedback”). I find that the latest collection ofpapers on motor control is as good for a laugh (or better) than a RobertBenchley collection. The best thing about it is that these people (the motorcontrollers; not Benchley) are SERIOUS.
Date: Tue Jan 12, 1993 11:03 am PST
Subject: S-R& PCT
[From Bill Powers (930112.0900)] Greg Williams (930112) —
> … where does a model for tracking which uses C,H, and T as itsvariables fit? And where does a model of gravitational attraction which usesdistance and mass (observationally proportional to weight) fit?
C, H, and T (and D) are the observable variables whose behavior we mustexplain, just as position, velocity, and acceleration are the observablevariables which Newton had to explain. The explanation goes beyond theobservable variables, in proposing entities like mass, or errorsensitivity.
Mass, remember, is not an observable variable (weight is not the same thingas mass), and neither is the universal constant of gravitation, nor the inverse-squarerelationship of force (acceleration for a free body) and distance. A pound offeathers and a pound of lead do NOT fall at the same rate. The attractionbetween nonspherical objects does NOT go as the inverse square of the distancebetween their centers. A cannonball does NOT travel in an ellipse with onefocus at the center of the Earth. The observations deny Newton’s laws. Newtonreplies, “Yes, but the underlying relationships are as I suppose. If youcalculate viscous friction, and integrate using my universal law over all theinfinitesimal particles of irregular objects, however imaginary those particlesmay be, you will see that the laws predict exactly what we observe.”
Suppose we have a simple control system with a loop gain of 1,000,000 and aslowing factor in the output function that is sufficient for stable operation.In this case, the true steady-state relationship between C and H is H = 1,000,000 * (C* -C). This is not, of course, what we observe. We observe that H = -Dand C = C*, as near as we can measure, give or take random noise andmeasurement error. If there are variations in C* we will see C varying in thesame way, but H will not vary a million times as much. H will vary only as muchas needed to cancel the effect of D on (C* -C). We do NOT, in general, see H varying a million times as much as C. Yet agenerative model in which the error sensitivity is one million explains theobservations.
> Still, I think Skinner’s “prematurity” warning still counts forsomething…
Skinner was denying the usefulness of models of the interior of theorganism at exactly the same time the principles of control theory were beingdeveloped. He defended his views against cybernetics and cognitive models asagainst any other proposals. He took the deviations of others’ views from hisown as prima facie evidence that the other views were wrong: no proof needed.Skinner’s main modes of argument were ridicule and assertion. He did not testhypotheses. He simply offered positive instances worded so as to support hisposition.
Suppose that Skinner had really believed, as he seemed to claim, thatmodels of the inner working of organisms might some day provide explanatoryprinciples not present in radical behaviorism. In that case, all hisexplanations of behavior in terms of external events and situations should havebeen appended with ” … or some cause working from inside the organism.”Obviously there was no such appendage: it would have made his bold assertionslook foolish. What Skinner believed, as far as I can see, must have been whatmany cognitive scientists believe today: that if you followed all more abstractexplanations down to their fundamental bases, the causes would eventually traceback to the environment. In other words, Skinner considered that he was onlystating in an approximate way what would some day be shown to be the onlyaccurate way. This was his lifelong faith.
> Actually, from what I’ve read, they actually claim that they DON’TWANT TO COME UP WITH AN EXPLANATION –ONLY PREDICTION/CONTROL. But the upshot is as you say, of course, and theycan’t get as close to their professed goal as they could with PCT models(which, as noted yet again above, might be very difficult to generate forcomplex situations).
The trouble with qualitative language is that you don’t get any idea ofproportions. To say that they can’t get as close to prediction as PCT can canleave the image of a footrace with PCT winning in a final burst at the finishline. With respect to prediction, PCT is crossing the finish line while theoperant approach still has one foot in the starting block.
Do you realize that there is no basis in the operant-conditioningmodel for predicting that there will be any behavior at all in a Skinner box?And that even if you admit as a prediction an extrapolation from previousexperience, this model can’t predict how much behavior there will be, if any?The reinforcement rate supposedly sustains the behavior rate. But thereinforcement rate depends on behavior, so unless you know in advance what thebehavior rate is going to be, you can’t say what the reinforcement rate willbe. Not being able to predict the reinforcement rate, you have no basis forpredicting any particular behavior rate. So there is NO PREDICTION ATALL.
The best that the operant approach can do is to describe what has alreadyhappened, and predict that what has happened will happen again. All of themathematical manipulations I have seen in the operant literature have beenmanipulations of algebraic identities; with only one equation to represent asituation requiring two or more equations for its complete description, that isall that can be done. It is not that the operant model predicts less well thanPCT. It does not predict AT ALL.
> Skinner’s extreme historical environmentalism and an extreme “moment-by-moment”mechanistic organismism need melding into a broader –and I think truer –picture.
As Dennis pointed out, “history” is not a causal mechanism. The past can’taffect the present any more than the future can. Everything that operates onbehavior is present now, or else it has no effect. The only way for the past toseem to operate in the present is through memory; and it’s only the currentcontents of memory, not what actually happened in the past, that has an effectNOW.
This is a fundamental principle of all the hard sciences. The history of avariable is irrelevant; the path by which it got to its present state(including derivatives) is of no consequence. Generative models work strictlyin present time. I don’t see any possibility for a merger here.
Avery.Andrews (930111.1905) —
> Reading around in the Osherson, ed. `foundations of cognitive science'(1989) I came across the claim that feedback is too slow to solve inversekinematic & dynamic problems for fast movements.
This is a good one for the collection of myths. In fact this is a doublewhammy: a myth based on a myth. In the first place, it’s not necessary to solvefor the inverse kinematics because a control system automatically does thatusing the environment as its own model. So it’s true that control systemsaren’t fast enough to solve the inverse kinematics without feedback: no neuralsystem is fast enough. But solving the inverse kinematic problem isn’tnecessary in any case, and control systems are certainly fast enough to do whatis necessary to control a dynamical system.
> Where can I read about why this claim is false or irrelevant (e.g.,true only for certain kinds of highly skilled movements that people practiceenough to make it plausible that they have elaborate feedforward schemesfor).
Look up analogue computer methods for solving second-orderdifferential equations. Korn and Korn is the only reference in my head, andit’s probably way out of date (Greg?). See also my arm model, which controls adynamical system without solving the inverse kinematic or dynamicequations.
I know of no situation in which literally solving the inverse kinematic anddynamic equations is a plausible explanation for behavior.
Best to all, Bill P.
Date: Wed Jan 13, 1993 7:51 pm PST
Subject: Feedback is too slow
[FROM Dennis Delprato (930113)]
Allegations that movements occur too rapidly for feedback to play a rolerepresent one of the major arguments against feedback control theories. Thecentral-peripheralissue goes back to the 19th century, and peripheral control was tied to”response-producedfeedback” –usually as a stimulus (!) for the next response. Thus, control theory as weknow it never really came up. Nonetheless, feedback control theory of themodern type (I acknowledge the major contributions of K. U. Smith and W. T.Powers for this) has been saddled with the near-ancientarguments used against its very distant relative.
From S. W. Keele (Attention and Human Performance, Pacific Palisades, CA:Goodyear Publishing Co., 1973), it appears that data interpreted in support ofthe allegation go as far back as Woodworth (1899, The accuracy of voluntarymovement, Psychol. Rev.). Lashley (1917, The accuracy of movement in theabsence of excitation from the moving organ, Amer. J. Physiol., v 43, 169-194)also cites Woodworth (1899). In this 1917 paper (well in advance of his much-cited1951 “The Problem of Serial Order” chapter) Lashley brings up themusician.
I found it interesting to be reminded of just how influential the setconstruct was. This is especially the case insofar as the set construct issimply an earlier version of today’s notions of cognitive control, motorprograms, schema, et al. The latter tell us not a bit more than did the ideaof set or determining tendency. Such is progress.
As far as feedback control theoretical arguments against the feedback istoo slow position, I have been able to find this addressed by T. J. Smith andK. U. Smith. This is in a preprint of a chapter entitled “Feedback-ControlMechanisms of Human Behavior” prepared for the Handbook of Human Factors/Ergonomics. I received the preprint approx. 1987; thus, the book must bepublished by now. In the preprint, the authors address our topic in Section184.108.40.206 (Control Theories of Fast Movement Coordinations).
They point out that it is virtually impossible to test the alternative tofeedback control –central brain programming–and that feedback control has been much evaluated, including with musicians.They find that feedback delay has major debilitating effects on performance tothe point that the highly -skilled musicians they used invariably refused to continue with the processafter brief exposures fearing permanent impairments. One of their mainreactions to the argument of feedback being too slow to play a role in skilledmovements is that if this is the case, delay and other distortations offeedback (e.g., displaced visual feedback) should not impair performance. Butdata from a wide variety of performances (musical, drawing, writing, and manymore) clearly show increased impairments with increases in experimentally-manipulateddistortations and delays of feedback.
They also bring up considerations that, to me, require elaboration to beconvincing: what they call predictive control and body movementtracking.
Date: Wed Jan 13, 1993 10:26 pm PST
Subject: feedback too slow
[Avery Andrews 930114.1652] (Dennis Delprato (930113))
Cool. Smith & Smith have a useful bad guy list, tho with nothing after1981, so it would still be good to know what Schmidt 1982 has to say.
The Smith & Smith article ref. is:
Smith, T.J. and K.U. Smith (1987) `Feedback and Control Mechanisms ofHuman Behavior’, in Gabriel Salvendy (ed) _Handbook of Human Factors_, Wiley,pp. 251-293,esp. 266-268.
Looking it over quickly, a possible story is this: given the speed of piano-playingmovements, it does seem plausible to me that a motor program is necessary forthis. But necessary does not imply sufficient, & what the Smith &Smith discussion (of the effects of delayed feedback) shows is that motorprograms are not sufficient. Maybe it’s time for Sesame Street to include somesome vs. all drills alongside of in vs. on, etc.
And these sorts of movements are really lousy guides to the nature ofroutine activity, since they are a form of behavior which nobody can achievewithout an inordinate amount of practice, and many can perhaps not achieve withany amount of practice.
Date: Tue Jan 12, 1993 5:08 pm PST
Subject: feedback to slow
[Avery Andrews 930112.1133] Bill Powers (930112.0900)
>> Where can I read about why this claim [feedback too slow] is falseor irrelevant (e.g., true only for certain kinds of highly skilled movementsthat people practice enough to make it plausible that they have elaboratefeedforward schemes for).
> Look up analogue computer methods for solving second-orderdifferential equations. Korn and Korn is the only reference in my head, andit’s probably way out of date (Greg?). See also my arm model, which controls adynamical system without solving the inverse kinematic or dynamicequations.
This sort of reading isn’t enough to make an impression on theestablishment. Academics are usually too busy, & maybe in some cases, toostupid, to actively extract the relevant implications from the stuff they read–everythinghas to be spelled out in ghastly, explicit detail, not left to lie in mathbooks, or buried in CSG-Larchives or C code. Students have more time, but they also can’t expected tohave very broad backgrounds, so again, the needed bits and pieces have to belaid out and put together slowly and carefully.
It needs to be documented in full
a) who first made the too slow claim, and what they were talking about whenand why.
b) wherein what was said is factually wrong.
c) wherein it is irrelevant to the general case.
Could something like this be published in Biological Cybernetics?
Remember that one the most important aspects of academic streetfighting isto make the opposition look stupid (as Chomsky did to Skinner with his reviewof Verbal Behavior), but to do this effectively you have to lay out the detailsin a way to be clearly accessible to prospective students, and practitioners ofneighboring disciplines.
Date: Thu Jan 14, 1993 10:09 am PST
Subject: Re: feedback to slow
[Martin Taylor 930114 12:00] (Avery Andrews 930112.1133)
>It needs to be documented in full
>a) who first made the too slow claim, and what they were talking aboutwhen and why.
>b) wherein what was said is factually wrong.
>c) wherein it is irrelevant to the general case.
> Could something like this be published in BiologicalCybernetics?
I doubt it could be published. Who made the claim was presumably not anelectronics person. Early in EE you get taught (or did a few decades ago)about a concept called Gain-BandwidthProduct. With positive feedback the apparent gain goes up and the bandwidth(and normally the delay) goes down. With negative feedback the reverse shouldbe expected to happen. At least that’s my very vague and distant memory, andwhat feeds my intuition about this issue.
Maybe my memory is all screwed up, but if not, there’s nothing to publishexcept a reference to standard textbooks.
Date: Thu Jan 14, 1993 4:50 pm PST
Subject: RE: Feedback is too slow
[From Bill Powers (920114.1400)] [Martin Taylor 930114 12:00]
> Who made the claim was presumably not an electronics person.
Too right. Much of this confusion can be traced back to earlycyberneticians, almost none of whom had engineering knowledge of controlsystems. Psychologists tended to treat anyone with the label cyberneticist as atechnonerd, so when they decided to believe something a cyberneticist wrotethey invested it with full authority. That’s how a lot of myths about controltheory got into circulation.
Best, Bill P.
Date: Thu Jan 14, 1993 6:10 pm PST
Subject: Feedback too slow
[FROM: Dennis Delprato (930114)] >Avery Andrews 930114.1652
> Cool. Smith & Smith have a useful bad guy list, tho with nothingafter 1981, so it would still be good to know what Schmidt 1982 has tosay.
Does this mean you don’t have access to R. A. Schmidt’s Motor Control andLearning, referred to by Jordan & Rosenbaum (in Posner, Ed., Foundations ofCog. Sci.? I can probably find a copy in a few days and promise to try if thismight help get you to write a paper on this issue. You mentioned BiologicalCybernetics as an outlet. Given that this is one of the oldest (mis?)conceptions in psychology, I imagine that a thorough, well-documentedand well-reasonedpaper would be given some consideration by several editors. Hope you get achance to see what Woodworth (1899) did and concluded –and let us know.
> The Smith & Smith article ref. is:
> Smith, T.J. and K.U. Smith (1987) `Feedback and Control Mechanisms ofHuman Behavior’, in Gabriel Salvendy (ed) _Handbook of Human Factors_, Wiley,pp. 251-293,esp. 266-268.
> Looking it over quickly, a possible story is this: given the speed ofpiano-playingmovements, it does seem plausible to me that a motor program is necessary forthis. But necessary does not imply sufficient, & what the Smith &Smith discussion (of the effects of delayed feedback) shows is that motorprograms are not sufficient. Maybe it’s time for Sesame Street to include somesome vs. all drills alongside of in vs. on, etc.
I am not an authority in the motor skills area, but it seems to me thattheorists have overlooked the possibility that what’s happened with thedevelopment of very rapid movements is that larger and largerphysical/physiological units have become functional psychological responseunits or patterns. Thus, the psychologist need not explain control of theunits themselves –I am quite certain physiologist Tom Smith would hold that physiological dataindicate nothing but feedback control processes at the physiological level.Level of description/explanation is a relevant consideration always. From apsychological level, what appears ballistic is not so when described from thephysiological perspective. Up to the present, when researchers have observedresponse patterns, they have assumed that the patterns needed explanation froma psychological perspective, hence suggestions of mechanistic chaining(behaviorists) or mental organization and so on (cognitivists). Neither sideconsidered that the patterns (organizations of physical/ physiological events)were simply givens, with the only needed explanation being their development(principles of how re-organizationtakes place). Note, for example, how Lashley (1951) argues for “motorpatterns” that “require the postulation of some central nervous mechanism whichfires with some predetermined…” (p. 123). So they might. But would this beequivalent to “cognitive control?” This is the common reasoning.
R. W. Pew (1974, chapter in B. H. Kantowitz’s Human InformationProcessing: Tutorials in Performance and Cognition) seems to take a move in thedirection I suggest. To a lesser extent, also, perhaps does Glencross(1977).
> And these sorts of movements are really lousy guides to the nature ofroutine activity, since they are a form of behavior which nobody can achievewithout an inordinate amount of practice, and many can perhaps not achieve withany amount of practice.
I agree. What about all the awful musicians and typists? Nevertheless,these movements are found and must be explained.
Dennis Delprato firstname.lastname@example.org
Date: Thu Jan 14, 1993 6:45 pm PST
Subject: feedback too slow; psychologists & cybernetics
[Avery Andrews 920104.1339] Dennis Delprato (930114)
>> Cool. Smith & Smith have a useful bad guy list, tho withnothing after 1981, so it would still be good to know what Schmidt 1982 has tosay.
> Does this mean you don’t have access to R. A. Schmidt’s Motor Controland Learning, referred to by Jordan & Rosenbaum (in Posner, Ed.,Foundations of Cog. Sci.? I can probably find a
Yes. This would be very helpful. I can’t promise a good product realsoon, since I’m feeling overextended & have to do some linguistics soon,but I do promise to try to try, as it were.
I agree entirely that levels is the key here. Presumably what there is inpiano playing is systems that set a series of reference levels in sequence,without waiting for evidence that one has been achieved in order to set thenext, with the details in charge of lower-levelsystems (are the spinal reflex loops fast enough to be useful in controllingpiano playing, or are their effects actually disturbances that have to be pre-compensatedfor?). But many higher-levelaspects of the performance then require perception of what has happened already(I’m a firm believer in the existence and importance of response-chainingin everyday life).
In K. VanLehn’s `Problem Solving and Cognitive Skill Acquisition’ in thePosner volume, p. 555, there’s an interesting statement about automization ofskills:
If exactly the same task is practiced for hundreds of trials, it can beautomatized, that is, it will be very rapid, cease to interfere with concurrenttasks, and run to completion once started even if the subject tries to stop it.If the task varies beyond certain limits during training, however, evenhundreds of practice trials to not suffice for automatization (Schneider andShiffrin 1977, Schiffrin and Sneider (1977).
the citations are:
1) Controlled and automatic human information processing:I. Detection,search and attention. Psychological Review 84:1-66.
2) Controlled and automatic human information processing:II. Perceptuallearning, automatic attending, and a general theory.
Psych. Rev. 84:127-190
(Lest people get false ideas about my ability to cover literature, the bookliterally just fell open at that page, so I read the paragraph at the top.Pennies from Heaven.)
So my story would be that there are automatic routines, which arepresumably normally very short, but with extensive practice can be made larger.And may be the skill acquisition literature can provide some actual evidenceabout what kinds of disturbances these lower level units can handle (assumingthat that’s what `varying the task’ involves).
Bill Powers (930114), psychologists & cyberneticians
I found some likely-lookingreferences to feedback bungles in Quantitative Analysis of Purposive Behavior,but who actually commits the Input Error?
Date: Fri Jan 15, 1993 9:31 am PST
Subject: Re: feedback too slow
[Chris Malcolm] Avery Andrews writes:
> I came across the claim that feedback is to slow to solve inversekinematic & dynamic problems for fast movements. Where can I read aboutwhy this claim is false or irrelevant?
It’s true for assembly (arm-handtype) robots. See any intro robotics textbook. One of the reasons why assemblyrobots can’t compete with human assembly speeds is that solving the inversekinematics at the speed of fast human arm movements is just on the brink ofbeing too expensive computationally. Inverse dynamics is orders of magnitudeworse. But of course we are talking here about algorithmic solutions based onmathematical models, and that is not the only way of solving these things (e.g.Churchland’s crab). In the end you come up against imagination failure, i.e.,”we can’t think of a way of doing this, so we think it is probablyimpossible.”
Date: Mon Jan 18, 1993 9:03 am PST
Subject: Devil’s Bibliography
[From Rick Marken (930118.0800)]
Here are some entries for the “Devil’s Bibliography”
1. Sheridan, T. B. and Ferrell, W.R. (1974) Man-machinesystems, Cambridge, MIT Press
Figure 9.1 Shows “reference input variable” as a variable outside the humanoperator where it is combined with an output variable to produce the inut tothe subject, e(t), the “error variable” which causes (via the operator functionY(H)) the “control variable” that influences the output of the “plant”.
Figure 11.10 (on quasi-linearmodel of a car driver) shows the “desired path” (of the car) as an INPUT to thedriver.
There are probably more but I’m going to look for a newer source. Ah,here’s one:
2. Huchingson, R. D. (1981) New horizons for human factors in design, NewYork: McGraw Hill
Figure 5.15 (p 181) shows a “command” signal being combined with systemoutput to produce the display variable that is the input to the “man” componentof the system. The “man” box converts the display input into an output that isthe “input” to the machine that produces the output that ultimately becomes thedisplay input.
Getting even more recent (and with no obvious signs of anyone catching on)we find:
3. Boff, K.R., Kaufman, L and Thomas, J.P (1986) Handbook of perception andhuman performance: vol.II, Cognitive processes and performance. New York:Wiley
Chapter 39 (by C. Wickens) is chock full of the “input blunder” and “manmachine blunder”. Particularly clear examples are:
Figure 39.1 which shows the error signal (e(t)) as the input to the humanoperator.
Figure 39.17 which shows the “optimal control model” as a sequence oftransformations of input yielding a “control” output; the secular contributionof the operator is observation and motor noise –quite an operator, though not completely unlike some people I’ve met.
Figure 39.39 shows an input being turned into a “displayed error” andpresented to the operator who turns it into the output variable.
Finally, the latest source that was nearby:
4. Weiner, E. L. and Nagel, D. C (1988) Human factors in aviation. NewYork. Academic Press
In Chapter 11, “Pilot control” by Sheldon Baron we find Figure 11.4 whichshows altitude and pitch signals as inputs to the pilot “box” which transformsthese inputs, via the function Y(P), into outputs that enter the vehicle’smachinery.
I could probably find some articles from the 1990s by searching throughrecent issues of journals like “Human Factors” but I like the stuff in these”bibles” of “human performance” knowledge the best.
Best regards Rick
Date: Mon Jan 18, 1993 9:24 pm PST
Subject: Re: Devil’s Bibliography
[From Bill Powers (930118.1600)] Rick Marken (930118.0800) —
You spur me to add a few more entries to the “Devil’s Bibliography.”
Here is Warren S. McCulloch helping to form the myth that feedback systemsgo unstable when their gain exceeds unity:
McCulloch, W. S.; Finality and form; in Embodiments of Mind, (Cambridge,MIT Press, 1965) pp. 256-275.
When we change the magnitude of the quantity measured, a reflex may returnthe system toward, but not quite to, the original state, or it may overshootthat state. The ratio of the return to the change that occasioned it is calledthe _gain_ around the circuit. When the return is equal to the change thatoccasioned it, then the gain is one…..if the gain is greater than one at thenatural frequency of the reflex, fluctuations at that frequency begin and growuntil the limitations of the structure composing the path reduce the gain toone; then, at the level for which the gain has become one, both the measuredquantity and the reflex activity go on fluctuating.” (p. 267)
Even earlier than this confident mangling of closed-loopproperties, we have this:
McCulloch, W. S., Appendix I: Summary of the points of agreement reachedin the previous nine conferences on cybernetics. _Cybernetics_: circular causaland feedback mechanisms in biological and social systems. Transactions of theTenth Conference, April 22,23 and 24, 1953, Princeton, NJ. Josiah Macy, Jr.Foundation, 1955. LCN 51-33199.
The transmission of signals requires time, and gain depends on frequency;consequently, circuits inverse for some frequencies may be regenerative forothers. All become regenerative when gain exceeds one. Regeneration leads toextreme deviation or to schizogenic oscillation…” (p71)
This rules out negative feedback systems with loop gains higher than unity –in other words, all actual control systems that exist in organisms.
On explaining how control works:
Wiener drew a most illuminating comparison between the cerebellum and thecontrol devices of gun turrets, modern winches, and cranes. The function of thecerebellum and the control of those machines is, in each case, to precomputethe orders necessary for servomechanisms, and to bring to rest, at apreassigned position, a mass that has been put into motion which otherwise, forinertial reasons, would fall short of, or overshoot, the mark.” (p. 72)
Here we have the germ of the compute-then-executeapproach.
There is a certain kind of intellect which can, on hearing the merestsummary of an idea, immediately leap ahead to its most profound implicationsand applications, completely unaware, or unconcerned, that it has a superficialand mostly incorrect understanding of the idea.
Here’s another gaggle of myths, this time from W. Ross Ashby, in _AnIntroduction to Cybernetics (New York: Wiley, 1966 (third printing, copyright1963).
The basic formulation of s.11/4 assumed that the process of regulationwent through its successive stages in the following order:
1. A particular disturbance threatens at D;
2. it acts on R, which transforms it to a response;
3. the two values, of D and R, act on T _simultaneously_ to produce T’soutcome;
4. the outcome is a state of E, or affects E.” (p. 221)
E is an essential variable that is to be stabilized by the action of aregulator R, acting through an environmental function T which is fixed.Regulation is achieved when the effect of D on T is precisely cancelled by theresponse of the regulator R to D, also acting on T. It is assumed that Edepends on T and T alone, so there are no disturbances acting directly on Ethat can’t be sensed by the regulator.
If R and T are precisely calibrated and act with infinite precision, thenperfect regulation is possible –but not otherwise. Ashby tended to overlook the question of precision, largelybecause in examples he tended to use small integers or decimal fractionsaccurate to one decimal place to represent the variables. As a result hegreatly overestimated the capacities of compensating systems, and therefore, bycomparison, greatly underestimated the capacities of control systems.
_Regulation by error._ A well-knownregulator that cannot react directly to the original disturbance D is thethermostat-controlled water bath, which is unable to say “I see someone coming with a coldflask that is to be immersed in me –I must act now.” On the contrary, the regulator gets no information about thedisturbance until the temperature of the water (E) actually begins to drop. Andthe same limitation applies to the other possible disturbances, such as theapproach of a patch of sunlight that will warm it, or the leaving open of adoor that will bring a draught to cool it.” (p. 222).
Note the implication that a compensating regulator might exist which, onseeing someone approach with a flask, could deduce that it contains cold waterand is about to be immersed in the bath. Note also the unspoken assumption thatmerely from qualitative knowledge about a flask of cold water, a patch ofsunlight, or a potential draught through an open door, the regulator could beprepared to act quantitatively: to add heat to the bath that would exactlycompensate for the cooling from the water in the flask or the evaporation dueto the draught, or cooling just sufficient to prevent any rise in thetemperature of the bath. From qualitative knowledge of the disturbance, theregulator somehow achieves exact quantitative compensation for the quantitativeeffects of the disturbance. If, of course, such a thing were possible, thecompensator would be much superior to any form of feedback controller. But sucha thing is not remotely possible.
After doing through a series of diagrams, Ashby finally diagrams the trueerror-drivencontrol system:
… we have the basic form of the simple ‘error-controlledservomechanism’ or ‘closed-loopregulator,’ with its well-knownfeedback from E to R.” (p. 223)
The diagram is D ->T ->E
Now we get to a whole fountain of misinformation about control systems, aseries of deductions that is just close enough to reality to be convincing, andjust far enough from it to be utter nonsense.
A fundamental property of the error-controlledregulator is that _it cannot be perfect_ in the sense of S.11/3″ (p.223)
He then goes through a “formal proof” using the Law of Requisite Variety toconclude
It is easily shown that with these conditions _E’s variety will be aslarge as D’s_ –i.e., R can achieve no regulation, no matter how R is constructed (i.e., nomatter what transformation is used to turn E’s value into an R-value).”
If the formal proof is not required, a simpler line of reasoning can showwhy this must be so. As we saw, R gets its information through T and E. Supposethat R is regulating successfully, then this would imply that the variety of Eis reduced below that of D –perhaps even to zero. This very reduction makes the channel
D ->T ->E
to have a lessened capacity; _if E should be held quite constant then thechannel is quite blocked_. So the more successful R is in keeping E constant,the more does R block the channel by which it is receiving its necessaryinformation. Clearly, any success by R can at best be partial.” (p. 223-224)
This argument has apparently convinced many cyberneticists and others thatthe Law of Requisite Variety is more general than the principles of control,and in fact shows that control systems are poor second cousins to compensatorswhen it comes to the ability to maintain essential variables constant againstdisturbance.
In fact this argument shows how utterly useless the Law of RequisiteVariety is for reaching any correct conclusion about control systems.
Having swept through this dizzying exercise in proving a falsehood, Ashbythen grudgingly allows feedback control to creep humbly back into thepicture:
Fortunately, in many cases complete regulation is not necessary. So far,we have rather assumed that the states of the essential variables E weresharply divided into “normal” … and “lethal”, so occurrence of the”undesirable” states was wholly incompatible with regulation. It often happens,however, that the system shows continuity, so that the states of the essentialvariables lie a long a scale of undesirability. Thus a land animal can passthrough many degrees of dehydration before dying of thirst; and a suitablereverse from half way along the scale may justly be called “regulatory” if itsaves the animal’s life, though it may not have saved the animal fromdiscomfort. “
Note the gratuitous “half way along the scale.”
Thus the presence of continuity makes possible a regulation that, thoughnot perfect, is of the greatest practical importance. Small errors are allowedto occur; then, by giving their information to R, they make possible aregulation against great errors. This is the basic theory, in terms ofcommunication, of the simple feedback regulator.” (p. 224)
The argument then veers off into “Markovian machines” and Markovian –stochastic –regulation. This is billed as the most important and far-reachingapplication of the error-controlledregulator.
Note how the argument relies on qualitative statements to reachquantitative conclusions. It is perfectly true that if a compensating regulatoraffects T equally and oppositely to the effect of D, E will not be affected atall. But by that same argument, to the extent that R does not have perfectinformation about D (and about the nature of the connection from D to T andfrom R to T), T will not be affected equally and oppositely, and thus to theextent of the imperfection, E will not be perfectly regulated. Furthermore, ifthere is any disturbance at all that is NOT detected by R (for example, adisturbance that acts directly on E), the effects of that disturbance will notbe compensated at all. If R does not compensate for all nonlinearities and time-functionsin the connection from D to T, compensation will not occur. When the processesinvolved are thought of as real physical processes in a real environment, theidealized assumptions behind the compensatory regulator are easily seen to beunrealistic –they predict regulation that is far, far better than any that could actually beachieved in this way.
Note also how the qualitative concept that error-regulatedcontrol must be imperfect is used to imply that it must be _more imperfect thancompensatory regulation_. This non sequitur has appeared in the literature overand over ever since Ashby. In his earlier book Ashby was still toying with truefeedback control and continuous systems; the appendix is heaped with ratheraimless mathematics that is oriented in that direction. But in this secondbook, Ashby shows that he never understood how an “error-controlled”regulator works. He didn’t know that the “imperfection” inherent in suchsystems can be reduced to levels of error far smaller than the error-reductionsthat any real compensating system could achieve –smaller by orders of magnitude, in many cases, particularly cases involvinghuman behavioral systems.
Ashby’s entire line of reasoning about feedback control in _An introductionto cybernetics_ is spurious. Yet Ashby has been revered in cybernetics andassociated fields for 40 years as a deep thinker and a pioneer. His Law ofRequisite Variety has nothing at all useful to say about control systems –and in fact led Ashby to a completely false conclusion about them — yet it is still cited as a piece of fundamental thinking. Whether Ashbyoriginated these misconceptions or simply picked them up from others I don’tknow. One thing is certain: he did not get them from an understanding of theprinciples of control.
Here’s a little test.
I have 200 pounds of ice cubes, and you have 50 gallons of boiling water.Desired: a nice tub of water for a bath. I get to throw in the ice cubes (youcan see exactly how may I throw in); you get to pour in the boiling water. Asyou see me disturbing the bath with ice-cubes,you estimate how much boiling water to pour in to arrive at a bath of the righttemperature. When I have exhausted my ice cubes, you finish the process byadding more boiling water in the amount you think is necessary.
As an alternative, I will let you see a thermometer in the tub, but willnot let you see how many ice cubes I am throwing in. You must base youradditions of boiling water entirely on the thermometer reading.
Whichever method of filling the tub you elect, when the process is finishedyou must then step into the tub and immediately sit down in it.
Which method would you choose: compensating for known disturbances, orbasing your action on perception of the state of the essential variable withoutknowing what the disturbances are?
Best, Bill P.
Date: Mon Jan 18, 1993 10:14 pm PST
Subject: Re: Devil’s Bibliography
Who shall be responsible for looking after it? I’ll try to integrate, butall contributors should keep ahold of their own stuff.
My feeling is that there should be a bad guy list (McCulloch, Arbib, atleast on the basis of his Hand. of Phys. diagrams (when I reread theaccompanying discussion, it started looking even worse, since he was manglingconcepts presented quite clearly and coherently by Houk and Rymer in theirHandbook article), for real trash by famous people, and a good guy list, forstuff that is basically sensible (Camhi, Houk & Rymer from what I’ve seenso far), although arguably hindered by a bad heritage.
Hmm. I think I remember trying to make sense out of the Ashby book when Iwas in school …
Date: Tue Jan 19, 1993 12:24 am PST
Subject: Devils Bibliography
[Avery Andrews 930119.1919] Here’s one from me:
Arbib, M. (1981) `Perceptual Structures and Distributed Control’, in_Handbook of Physiology_, Motor Control II.
Figure 11 on pg. 1446 is supposed to represent a `feedforward’ systemoperating alternately with a `feedback’ system, where the `feedforward’ systemoperates to correct large errors, and the `feedback’ to correct small ones.But it is clear that the `feedback’ system is just a `gross’ feedback system -one with low gain, but high efficacy to correct crude errors.
Figure 12 (pg. 1447) is supposed to represent `coactivated’ systems, wherethe `feedforward’ system injects signals concurrently with the `feedback’ one.But this is also wrong: the `feedforward’ system has only one input line (from`desired output’ = reference signal), and since it knows nothing about what is,it can’t achieve any useful effect. It needs (at least) another input line,representing info environment from which undesirable influences on the outputcan be predicted.
The accompanying text does not improve things. On pg. 1445 feedforward isdescribed by paraphrasing a (clear and cogent) exposition by Houk and Rymer(1981), who describe feedforward as monitoring the environment for factors thatwill create predicatable disturbances to the controlled quantity, and injectingappropriate signals to counter the deviations before they actually happen.Later we get:
“Here we extend the sense of feedforward to include a strategy thatgenerates control signals to rapidly bridge large discrepancies in desiredoutput at too great a velocity for long-latencyfeedback paths to have any effect.” which seems confused. Then alpha-gammaefferent coactivation is presented as an example of the coactiviation model ofFig 12, which seems pretty dubious (a non-dubiousone, using the Houk & Rymer concept of feedforward, would be turning thesteering-wheelin response to perceived curvature in the upcoming road, before there is anyvisible drift of the car from a desirable position on the road).
Date: Tue Jan 19, 1993 12:27 am PST
Subject: feedback too slow
[Avery Andrews 930119.1922]
Maybe another contributing factor to the `feedback too slow’ story is thework by Rack and others (who seem to know what they’re doing) arguing that thespinal reflexes are too slow and have too little gain to be much use incompensating for short-termdisturbances.
Hopefully this doesn’t imply that they’re not useful for dynamics -it is perhaps significant that amphibians, who lack the dual alpha-gammaefferent system, are not noted for their acrobatic abilities.
Date: Tue Jan 19, 1993 4:58 am PST
Subject: Schmidt quotes
From Greg Williams (30119)
The following excerpts are from Richard A. Schmidt, MOTOR CONTROL ANDLEARNING: A BEHAVIORAL EMPHASIS, Human Kinetics Publishers, Champaign,Illinois, 1982 (dedicated to “former mentors” Jack A. Adams and Franklin M.Henry).
[diagram on page 204 shows “closed-loopmodel for movement control” with (surprise!) the reference signal andcomparator inside the person]
205 -“The idea is that the system [diagrammed on page 204] can ‘compute’ theexpected nature of… sensations in the form of a reference and can compare thefeedback it receives on a particular trial with the feedback it expects toreceive…. Closed-loopmodels… are thought of in essentially two ways. First, they provide a basisfor knowing if a movement produced is correct or not…. A second way…concerns the control of ongoing movements…”
206 -“The closed-loopmodel… has been very effective and useful for certain kinds of responses….the model seems to have considerable appeal for movements in which something isregulated at some constant value… These movements are called TRACKINGresponses… The most important generalization from… [tracking] research isthat if the models are used in computer and/or mechanical SIMULATIONS of thehuman in which the computer or mechanical device is controlled in waysanalogous to those in… [the diagram on page 204], these nonliving devices’come alive’ to behave in ways nearly indistinguishable from their humancounterparts. By proper adjustment of certain mathematical or electronicelements in the devices (called parameters), the system can be made to showmany human characteristics, and they track with essentially the same levels oferror…. The large body of experimental literature suggests that because thehuman can be mimicked so well by computers that use closed-loopmechanisms such as shown in… [the diagram on page 204], the human in these[tracking] situations can be regarded as a closed-loopcontrol system, responding essentially by analyzing the feedback producedagainst the reference of correctness and issuing corrections. The evidence doesnot prove that humans actually track this way, but the agreement betweentheoretical predictions and data is very strong, and alternative theoriescannot boast of similar success.”
210 -“Engineers can design robots and other machines to behave… using what theycall POINT-TO-POINTCOMPUTATION methods. The position of the limb at each point in space and ateach time in the movement is represented by a reference for correctness, andthe system can be made to track this set of positions across time to produce anaction with a particular form. But the system must be very ‘smart,’ and it mustprocess information very rapidly, even for the simplest of movements. All ofthese references for correctness must be stored somewhere, and each of thepoints will be different if the movement begins from a slightly different placeor if it is to take a slightly different pathway through space.
Engineers have generally found that these methods are very inefficient formachine (robot) control, which has led many motor behavior researchers (seeKelso, Holt, Kugler, & Turvey, [in G.E. Stelmach & J. Requin, eds.,TUTORIALS IN MOTOR BEHAVIOR,] 1980; Kugler, Kelso, & Turvey, [in same,]1980) away from these kinds of control processes to explain humanskills….
A compromise position is that only certain positions in the movements arerepresented by references of correctness. One viewpoint is that feedback fromthe movement when it is at its endpoint is checked against a reference forcorrectness and that subsequent corrections are initiated to move the limb tothe proper position. These views of motor control hold that the limb is more orless ‘thrown’ in the direction of the endpoint by some kind of open-loopcontrol and that then the limb ‘homes in on’ the target by closed-loopcontrol.”
211 -“… the information-processingmechanisms, which lie at the very heart of the closed-loopsystem shown… [on page 204], require a great deal of TIME AND ATTENTION forstimuli to be processed to yield a response…. with rapid actions sufficienttime is not available for the system to (a) generate an error, (b) detect theerror, (c) determine the correction, (d) initiate the correction, and (e)correct the movement before a rapid movement is completed. Muhammad Ali’s leftjab is a good example. The movement itself is about 40 msec; yet, according toour estimates [made earlier in the book on the basis of movement-correctionexperiments], detecting an aiming error and correcting it during the sameresponse should require about 150 to 200 msec –the time necessary to complete the activities of the stages of informationprocessing…. Another problem is that the time between two responses that arenot grouped (see… [M.C. Smith, in W.G. Koster, ed., ATTENTION AND PERFORMANCEII, 1969]) was found to be essentially 190 msec. Therefore, the time betweensuccessive corrections (because corrections are responses) should be similarlyseparated. This information has serious consequences for models of human limbcontrol that demand a large number of attention-basedcorrections in a very short period of time. Point-to-pointcomputation models have this basic problem when human performance isconsidered.”
212 -“What about the possibility that the central nervous system contains closed-loopmechanisms that do not require any attention?
215 -“… experiments like Dewhurst’s ([IEEE TRANS. BIOMED. ENGR. 14, 167-171] 1967) show that corrections for suddenly presented changes in position canbe initiated far more rapidly than the earlier 200-msecestimates, with correction latencies being from 30 to 80 msec in the variousinvestigations that have been done. This kind of result suggests that theinformation-processing stages are not involved in these corrections, as the stages requiretoo much time for processing.”
243 -“On strict experimental grounds [deafferentation studies], the evidence doesnot really show that the [open-loop]response-chaininghypothesis is incorrect, although it strongly suggests it. A more reasonablehypothesis would be that feedback is not essential for the production of atleast some movements, although it is likely that feedback provides increasedflexibility and improved fine movement control. The possibility remains thatunder some conditions or for certain kinds of skills the response-chaininghypothesis might be correct, but it seems fair to say that it is not correct ingeneral.”
249 -“Henry and Harrison (PERCEPT. MOT. SKILLS 13, 351-354]1961) asked subjects to begin with a finger on a key located by their hip andat a ‘go’ signal to move forward-upwardto trip a string in front of their right shoulder. They were to do this asquickly as possible. The simple RT in this situation was 214 msec on theaverage, and the movement time was almost the same, at 199 msec. Sometimes asecond light would come on indicating that the subject should avoid trippingthe string or at least begin to slow the limb as quickly as possible. The’stop’ signal could come on at one of four points: 110, 190, 270, and 350 msecafter the ‘go’ signal….
250 -Henry and Harrison measured the time to begin to decelerate the limb after a’stop’ signal.
Their results are simple. Only when the ‘stop’ signal was given at the 110-msec location was there a tendency for the subjects to start to slow themovement before it had been completed. But the more interesting feature ofthese data is the subject’s response in the 190-mseccondition. Notice here that the ‘stop’ signal came on 24 msec BEFORE themovement even started, and yet the movement was carried out withoutinterruption….
If the information-processingstages are too slow to be involved in details of a particular action…, thenthe question is: What does produce these patterns of action? The best theory tohave been proposed at this point is that these patterns are PREPROGRAMMED,structured in advance, and run off as a unit without much possibility ofmodification from events in the environment. To me, this is one of thestrongest lines of evidence available that movements are controlled by motorprograms.”
There is more, but I’m tired of typing for now. If this is useful, let meknow
and I’ll probably post another installment.
As ever, Greg
Date: Tue Jan 19, 1993 6:17 am PST
Subject: Re: Schmidt quotes
[From: Bruce Nevin (930119 08:44:23)] (Greg Williams (30119)
Controlling for the stop signal I assume is at a higher level of thehierarchy, and the delay corresponds to that in the portable demo described inBCP (extended arm is raised before lowering it in response to signal).
Isn’t some of their time problem due to thinking of the loop sequentially(trial and then error and then correction in another trial) rather thanconcurrently and continuously?
Date: Tue Jan 19, 1993 9:13 am PST
Subject: Devil’s Biblio/Schmidt
[From Rick Marken (930119.0800)]
I’m glad to see the contributions to the “Devil’s Bibliography”. Avery –keep posting candidate entries whenever you find them. They won’t get lost.Greg (as you can probably tell by now) is an excellent archivist andpublisher. Maybe someday the “Devil’s Bibliography” entries will be compiled inan issue (or two) of Closed Loop.
Greg Williams (30119) —
Thanks for the excepts from Schmidt. You ask:
> There is more, but I’m tired of typing for now. If this is useful, letme know and I’ll probably post another installment.
Yes. It’s EXTREMELY useful. Thank you for doing it and when your hand andarm return to normal please post some more. There are some real gems in whatyou posted already such as:
> 205 -“The idea is that the system [diagrammed on page 204] can ‘compute’ theexpected nature of… sensations in the form of a reference and can compare thefeedback it receives on a particular trial with the feedback it expects toreceive
Sounds right to me; of course, the comparison is part of a continuousprocess that makes the feedback sensations match the “expected” (reference)sensations. But the statement is consistent with (at least part of) PCT.
After describing how well closed loop models mimic behavior, Schmidt makesthe following, remarkable statement:
> The evidence does not prove that humans actually track this way, butthe agreement between theoretical predictions and data is very strong, andalternative theories cannot boast of similar success.”
So why the apparent eagerness to abandon closed loop models in favor ofother schemes? The following paragraph seems to hold a clue –it shows that there is a strong desire to maintain the “information processing”concept of the nervous system (stimulus input –>processing–>responseoutput). I believe this is the result of an unconscious assumption that the”causal model” that underlies the information processing view MUST be correct –closed loop feedback or not. This paragraph is a candidate for the frontpieceof the Devil’s Bibliography:
> 211 -“… the information-processingmechanisms, which lie at the very heart of the closed-loopsystem shown… [on page 204], require a great deal of TIME AND ATTENTION forstimuli to be processed to yield a response…. with rapid actions sufficienttime is not available for the system to (a) generate an error, (b) detect theerror, (c) determine the correction, (d) initiate the correction, and (e)correct the movement before a rapid movement is completed. Muhammad Ali’s leftjab is a good example. The movement itself is about 40 msec; yet, according toour estimates [made earlier in the book on the basis of movement-correctionexperiments], detecting an aiming error and correcting it during the sameresponse should require about 150 to 200 msec –the time necessary to complete the activities of the stages of informationprocessing.
The idea that this is all part of a continuous LOOP is lost in the dust ofthe causal (input-output)view of behavior.
Thanks again for all the work on this Greg.
Date: Tue Jan 19, 1993 9:36 am PST
Subject: Re: Devil’s Bibliography
[From Bill Powers (930119.0900)]
Today is JAMES WATT’S BIRTHDAY.
Ft. Lewis’ computer is down again; the Fine Arts building collapsed underthe weight of a new heavy slow and this has probably upset everything, althoughthe computer is not in that building. So I might as well go on with the Devil’sBibliography.
This is Myerson, J. and Miezin, F, M.; The kinetics of choice: an operantanalysis. Psychological Review Vol. 87, No. 2, 160-174(1980). Quotations are placed between pairs of dashed lines.
The results of a large number of experiments on concurrent schedules arewell described by the matching law (Herrnstein, 1961),
(1) B1/(B1+B2) = R1/(R1+R2)
where Bn is the rate of response or amount of time allocated to somebehavior _n_ and Rn is the obtained rate of reinforcement for this behavior.(p. 161)
This equation shows a relationship between behavior rate and reinforcementrate. Without altering the algebraic relationship, we can simplify it:
1 multiply by denominators:
B1*R1 + B1*R2 = B1*R1 + B2*R1
B1*R2 = B2*R1
2. B1/R1 = B2/R2, or B1/B2 = R1/R2
The first form of 2. states that the ratio of behavior rate toreinforcement rate on each schedule is the same. This will be trivially truefor any pair of identical fixed-ratioschedules, because Bn/Rn is merely the number of presses required to produceone reinforcement. Equation 2 states that this number is the same on bothschedules. Note that the “rate” aspect drops out because both numerator anddenominator contain inverse seconds as units.
However, the result will be wrong for any pair of fixed-ratioschedules in which the number of presses per reinforcement is not the same forboth choices. Whether the prediction is _judged_ wrong will depend on how farthe prediction must be from the data while the data are still considered “well-described.”There seems to be no reason to believe that the results will be any better forvariable ratio or for any interval schedules.
In the two-alternativecase, reinforcing a response for which preference is P1 = B1/(B1+B2) decreasespreference for the alternative, P2. The rate at which P2 decreases will beproportional to the rate R1, at which the first response is reinforced.Thus
(2) dP2/dt = -kR1P1
where the constant k governs the proportion of change perreinforcement.
Because preference is defined as _relative probability_, P1 + P2 = 1, anddecreasing the preference for one alternative must therefore increase thepreference for the other. Therefore, the decrease in P2 in Equation 2 impliesan equal and opposite reaction, a compensatory increase in P1:
(3) dP1/dt = +kR1P2 (p. 162)
Probability has nothing to do with it. If Pn = Bn/(B1+B2), then
P1+P2 = 1 because B1/(B1+B2) + B2/(B1+B2) = (B1+B2)/B1+B2) = 1.
If, however, in addition to reinforcing one response at rate R1, we alsoreinforce the other response at rate R2, then equation 3 must be supplementedto include the proportional decrease in P1 produced by reinforcement of thesecond response.
Note that R1 and R2 have suddenly become manipulated independentvariables.
Moreover, Equation 2 must be supplemented to include the compensatoryincrease in P2 that comes from decreasing P1. Therefore the complete system isdescribed by
(4a) dP1/dt = kr1P2 -kR2P1
(4b) dP2/dt = kR23P1 -kR1P2
This system is represented diagrammatically in Fig. 1a. Atequilibrium,
dP1/dt = dP2/dt = 0,
Whereupon Equations 4a and 4b both lead to
R1/R2 = P1/P2 = ———–= B1/B2
which is algebraically equivalent to Equation 1 (Baum & Rachlin, 1969).(p. 162)
Note that Equations 4a and 4b are not a “system” of equations. The twoequations are linearly dependent and are just two ways of writing the samerelationship.
So M&M have managed to run this series of permutations of the originalequation around to its starting point, which is a statement that the ratio ofbehaviors to responses is the same on both choices (even if the authors havefailed to see that this is what it states).
The general kinetic model just described is a hypothesis concerning thefunctional relation between the reinforcement input to the organism and thebehavioral output. (p. 163)
It is no such thing. In all operant experiments, reinforcement is afunction of behavioral output; that is what can be observed. Behavior is theinput to the scheduling apparatus; reinforcement is the output of theapparatus. There may be a simultaneous dependency of behavioral output of theorganism on reinforcement input to the organism, but to describe that wouldrequire a second basic equation to be solved simultaneously with thefirst.
By plodding through the rest of the paper one can show that every step ofthe development is just as far from real system analysis as the initial part.The mathematics gets more and more involved as the paper proceeds, so itbecomes more and more tedious to show that each successive form is merely atransformation of the preceding forms, extending the tautology (and the self-contradictions) to ever greater complexity, and that all the “predictions” ofdynamics are simply curve-fittingof arbitrary mathematical forms to the data. The entire paper is an attempt tomake deductions about a system-environmentrelationship by taking into account only one of the two necessary functionalrelationships –and attributing that one incorrectly to the organism.
Powers (1973,1978) has contended that systems analysis represents atheoretical alternative to behaviorism. However, the present treatmentdemonstrates the compatibility of the two approaches and suggests that thetechniques of systems analysis have much to offer to the mathematical behaviortheorist. Powers’ analysis differs from ours in two important ways. First, hisis based on a control theory model. Control theory is a branch of systemsanalysis that assumes the existence of control [sic] variables or referencevalues [sic sic sic] with which inputs are compared (Milhorn, 1966). Althoughsuch hypothetical reference values have their uses, they are by no meansnecessary for the analysis of feedback systems (Milhorn, 1966; Rosen, 1970).Second, in opting for a “quasi-static”analysis (Powers, 1978), we believe that Powers has forsaken perhaps the mostimportant attribute of systems analysis, its ability to describe both transient-stateanalysis and equilibrium behavior. (p. 172).
No comment required in the present company.
Best to all, Bill P.
Date: Tue Jan 19, 1993 1:27 pm PST
Subject: Devils Bibliography: A Special Corner
[Rick Marken (920119.1200)] Avery Andrews (930119.1919)
> Figure 11 on pg. 1446 is supposed to represent a `feedforward’ systemoperating alternately with a `feedback’ system
I think there should be a special corner in the hell of the Devil’sBibliography for pronouncements about the effectiveness of feedforward.
Feedforward is just Ashby’s “compensating regulator” that Bill Powers(930118.1600) demolished rather uncomfortably with his newly coined “parable ofthe bathtub”. The corner of hell for feedforward zealots should have Powers’bathtub attended by two devils, one with 200 pounds of ice cubes, the otherwith 50 gallons of boiling water –AND NO THERMOMETER ANYWHERE.
> feedforward is described by paraphrasing a (clear and cogent)exposition by Houk and Rymer (1981)
Houk & Pymer’s description of feedforward may be cogent –but if they think feedforward actually works (produces control) then it’s thePowers bathtub corner for them too.
Date: Tue Jan 19, 1993 2:04 pm PST
Subject: devils bathtub
[Avery Andrews 930119.730] Rick Marken (920119.1200)
Houk & Rymer (and even Arbib) don’t suggest that feedforward can beeffective on its own -in their diagram it acts in concert with a feedback system (like turning thesteering wheel on the basis of the perceived curvature of the upcoming road,which can be done in accord with a formula posted by Bill during the steeringdiscussion a few months ago. So I’d put H&R on the goodguy list (afterall, Houk & Milhorn (1984) described the posture-controlcircuitry that Bill & Greg simulated a simplified version of).
(Greg Williams (30119))
Wow! The page 210-211quotes are really off the wall. It seems to me that this stuff can be used inarm documentation. E.g. Jordan & Rosenbaum dismiss feedback as too slowciting only Schmidt, but judging from these quotes he’s demonstrably soignorant that they seriously discredit themselves by doing this.
`Event-based’analysis seems to be a common thread running through many of these DB entries,& it’s interesting that that’s where Beth Preston’s Synthese paper fallsapart too, in my judgement. Also assuming that reaction times for voluntaryactivity have much to do with the existence of lower level closed-loopmechanisms. I think that a lot of the movement-controlschemes in Jordan & Rosenbaum are event-basedtoo, but I’m not sure, since the prose is fairly vague (or my math toofeeble).
But, it does seem like at the vicinity of 212 Schmidt might be gettingaround to making sense, but I don’t get a clear conception of what overallconclusions he’s coming to there.
Date: Tue Jan 19, 1993 2:37 pm PST
Subject: Re: Devil’s Bib comments
[From Bill Powers (930119.1430)] Greg Williams (930119) —
RE: Devil’s Bibliography
The excerpts from Schmidt make him sound like an almost-good-guy.Actually, a good guy who is looking for whatever there is in control theorythat seemed to him usable. What he’s missing is the concept of a hierarchy ofcontrol. It’s interesting that even in 1982 he is citing the engineers’ “point-to-point”method of controlling movements, and the fact that it is computationallyexpensive. This idea clearly influenced Kelso et. al. It would be interestingto see if we could find where this approach to movement started –probably WAY back. The idea that a simple analogue signal variation could drivea controlled movement with hardly any computation at all seems to have beenlost along with analogue computers.
Is there any way to check out something about his “former mentor,” Jack A.Adams? Back in ’57 or ’58, Bob Clark, the late Bob MacFarland, and I went tothe U of IL to give a seminar for O. Hobart Mowrer’s graduate students. Amongthem, I think, was Jack Adams. Mowrer was very taken with our concepts of”feedback theory” as we called it then (and as the Adams in the literaturerefers to it). Mowrer devoted a good part of chapter 7 of _Learning theory andthe symbolic process_ (New York: Wiley (1960)) to our ideas. If this is thesame Jack Adams, that would be interesting, because as far as I know he doesn’tcite us or Mowrer. If it isn’t the same Adams, it’s not so interesting.
I just computed the minimum acceleration of Ali’s fist, from the remembereddata that his jab moved about 6 inches. If this took 40 milliseconds as citedby Schmidt, Ali’s fist accelerated at 187.5 meters/sec^2, or 19 g’s. Obviouslyno control is possible during such a brief movement, but it’s easilyexplainable as resulting from a step-changein the position reference signal. The Little Man exhibits such speeds ofmovement when the kinesthetic reference signal is stepped abruptly to a newposition (I set the parameters to give about 100 millisecond movements, but themodel can go faster). If the reference position is set _beyond_ the target, asfighters are supposed to do, the transition time to hit the target can be evenless than 40 milliseconds even with the model set to go 95% of the way to thetarget in 100 milliseconds.
Bruce Nevin (930119.0844) —
> Controlling for the stop signal I assume is at a higher level of thehierarchy, and the delay corresponds to that in the portable demo described inBCP (extended arm is raised before lowering it in response to signal).
Yes, that’s what I would say, too.
> Isn’t some of their time problem due to thinking of the loopsequentially (trial and then error and then correction in another trial) ratherthan concurrently and continuously?
I agree that it is.
> Poul Andersen did this sort of thing with a General Semantics premisein the early ’50s (_The World of Null-A_f’rinstance).
Writing under the name of A. E. van Vogt? Actually I think World of Null-Acame out a good deal earlier than the ’50s –it was my introduction to Korzybski. I remember the fizz of intellectualexcitement when I realized that the fascinating quotes used for chapterheadings came from a REAL BOOK! I spent my later high-school years trying todevelop that “cortico-thalamicpause.” It greatly improved my ping-ponggame.
Best to all, Bill P.
Date: Tue Jan 19, 1993 2:47 pm PST
Subject: devils bathtub
[From Rick Marken (930119.1400)] Avery Andrews 930119.730
> Houk & Rymer (and even Arbib) don’t suggest that feedforward canbe effective on its own -in their diagram it acts in concert with a feedback system.
Oh, all right. They can have the tub in the PCT corner of heaven which hasa thermometer that shows current temp, derivative of temp and a “predictivedisplay” of up to 10 minutes of estimated future temp –constantly revised based on changing outputs (ice cubes, boiling water andatmospheric conditions) –the delux model. I agree that this kind of “feedforward” could be used toimprove control (the predicted information is still part of a closed controlloop –the word “feedforward” may be what is misleading; predicted future informationis just a present time perceptual input that can be a controlled variable –or part of a controlled variable –like any other controlled variable).
Date: Tue Jan 19, 1993 3:29 pm PST
[Avery Andrews 930129.0952]
It seems to me that the critical thing w.r.t. Schmidt is to figure out whataspects of it Jordan and Rosenbaum were thinking of as evidence for feedbackbeing too slow. I leaped to the conclusion that it was the 210-211stuff, but we better be careful about getting this right. By the way, I don’tnotice what I would call the `event-basedblunder’ on the list in QAPB. Does it have a standard name?
Presumably at the bottom of the pit in our inferno would be McCulloch, whoseems to be deeply implicated in the event blunder & subvertingunderstanding of continuous/analog computing in general.
Houk & Rymer on the other hand strike me as definitely being good guys.The main problem with their exposition is that, although it is very clear,their diagrams are not appropriate for psychology (they seem based on therequirements of chemical engineering): there;s a box labelled `controlledsystem’ with arrows coming out labelled `regulated variables’, with theeffectors, comparators, etc. outside the `controlled system’.
ref error forcing | Controlled | regulated vars.
—->Sum —->Amplifier —->| System |————->
+ ^ function ———— |
| – |
Looking at this, it’s basically the same as Bill’s diagrams, but insideout: the `Controlled system’ is the entire environment, the forcing functionis the effectors, etc.
Date: Tue Jan 19, 1993 9:07 pm PST
Subject: Bathtubs; feedforward
[From Bill Powers (930119.2200)]
If I used all the ice and you used all the boiling water, and the ice wasjust freezing, I think the bath would come to 84 F or 29 c –pretty tepid. That’s 200 pounds of ice and 800 pounds of boiling water. Anybodycheck that out? Of course you don’t know how far below freezing the ice cubeswere. One assumes sea level and distilled water.
RE: feedforward in steering a car, and ignore this if I’m repeatingmyself.
If you adjust the place on the road where you look, you can anticipate thecurves without feedforward. By looking at the road just far enough ahead of thecar, you will see a deviation in the road before the car gets to the curve.This distance has to be adjusted for the speed of the car, so higher-levelcontrol is involved. With just the right distance, reacting to the disturbancein the usual way will result in the commencement of a steering effort just whenit’s needed. Try it: it works. Of course there’s no way to prove that anyoneever did it this way before.
Best, Bill P.
Date: Wed Jan 20, 1993 5:22 am PST
Subject: More Schmidt quotes
From Greg Williams (930120)
I think the burgeoning DB + critique thereof could be put to good (ifmischievous?) use. Why not work up the quotes and challenges/reinterpretationsand send copies to lots (ca. 50?) of “biggies” in the “harder” (???) end ofestablishment psychology? An epistle from CSGnet to YOU: we have some problemswith the basic assumptions… might you be able to set us straight. Even if notone joins the net, I bet at least a few will have their eyes opened a bit tothe fact that they have been glossing over some difficulties. Anybody else upfor this? The dialogue needs to escape the confines of the net, in my opinion –without the necessity of submitting to a journal. I think we need ongoinginteractions with REAL “devils.”
OK, Rick, here’s some more. (You ARE a nice guy, despite what they sayabout you on the net.)
253 -“… Keele and Posner ([J. EXP. PSYCHOL. 77, 155-158]1968)…. trained their subjects to move a hand held stylus (like a pencil) toa small target a few centimeters away. Knowledge of results about MT [movementtimes] was provided, and subjects were trained to move in either 150, 250, 350,or 450 msec, in separate sessions. Then, when the subjects had perfected theMTs, on certain trials (unpredictable to the subjects) the experimenters wouldturn off the room lights when the subjects left the starting position, so thatthe remainder of the movement would have to be made in the dark. The logic wasthis. If, in the 250-mseccondition, for example, the lights-oncondition produced less movement error than the lights-offcondition, then the lights being on must have contributed to accuracy, and theimplication would be that vision was being used for movement control.Conversely, if no differences in accuracy were obtained between the lights-onand lights-offconditions, then the implication would be that vision was not being used in theresponse and that visual feedback processing time was greater than thecorresponding MT in these responses.
Keele and Posner (1968) measured the probability of missing the targetfor the four MT categories for the lights-onand the lights-offconditions… The most important finding is that, in the 150 msec MT condition(the actual MT was 190 msec), about as many target misses were recorded whenthe lights were on (68 per cent) as when the lights were off (69 per cent). Theauthors argued that when the MT was 190 msec or less, the vision of the hand ortarget did not contribute to movement control; thus, they argued that usingvisual feedback required at least 190 msec (one RT [reaction-time])to process. Also, as the MTs increased, an advantage began to develop forhaving the room lights on. When the MT was 260 msec, there were 47 per centmisses for the lights-oncondition and 58 per cent misses for the lights-offcondition. The lights-onadvantage increased steadily as the MT increased, so that with a 450 msec MT,the lights-offcondition had 47 per cent misses and the lights-oncondition had only 15 per cent….
254 -Hawkins et al. [ms. in prep.] performed a series of experiments much like thatof Keele and Posner, except that their subjects practiced for a block of trialswith vision always presented, and then a block of trials with vision alwayswithheld, thus eliminating the problem of having the vision unexpectedlyremoved. They plotted the total variability around the target… as a functionof the MT… The lights-oncondition was beneficial at all MTs, even for those as short as 100 msec! Thesedata suggest that vision, when its presence can be expected, can be used in farless time than the 190 msec that Keene and Posner suggested, perhaps withvisual processing times as short as 100 msec…. Another experiment showed thatthe advantage of the lights-oncondition vanished when the MTs were 80 msec…”
255 -“Taken together, these findings seem to provide a contradiction to the generalnotion that vision presented in the environment cannot influence a movementuntil all of the stages of information processing have been completed –that is, until about 190 msec have elapsed. How can the advantage of visionoccur when the MT is far shorter than this 190 msec? One answer is that visionin these situations is not processed like a suddenly presented stimulus towhich a response must be made, because more or less continuous vision of thehand and target occurs during the movement….
Another possibility is that vision ‘tunes’ the motor system foraccuracy… It is possible that corrections in the movement do not occur atall, but that vision prepares the spinal apparatus for the ongoing movement…These ideas are speculative…”
299 -“… it does not make sense to claim that motor programs operate withoutfeedback.”
As ever, Greg
Date: Wed Jan 20, 1993 9:51 am PST
Subject: More Devil’s Bibliography
[From Rick Marken (930120.0900)]
No collection of misconceptions about PCT (and general feedback control)would be complete without a quote or two from:
Fowler, C. and Turvey, M. T. (1978) Skill acquisition: An event approachwith special reference to searching for the optimum of a function of severalvariables. In G. Stelmach (Ed.) Information processing in motor control andlearning (pp. 1-39)New York: Academic Press
An entire section of this article is dedicated to Powers HPCT model ofbehavior. It starts on p. 26 and runs to p. 31. Their efforts to show thelimitations of the HPCT model are detailed, quantitative and completely wrong.Despite numerous demonstrations of the fallacies and misconceptions in theFowler/Turvey paper (as well as letters to Turvey –the main “bad guy –see below) explaining these problems) there has never been any reply fromFowler/Turvey and certainly never an apology for this terribly mischievous”trashing” of HPCT.
Here are a couple of quotes from the end of the section describing(incorrectly) HPCT.
” Quantitative knowledge of results must rarely be informative in ahierarchical closed-loopsystem because, typically, there is a one to many mapping between an errorsignal and the conditions that may have provoked it. We can conclude from that,perhaps, that the actor/perceiver is not appropriately characterized as ahierarchy of control systems at least when he is performing tasks in which hemust exploit the abstract information putatively extracted by the superordinatelevels of the system”. p 31
This statement is the typical mixture of some correct statements and someincorrect assertions that is the hobgoblin of PCT reviews. Yes, there is a oneto many mapping of error to outputs in a hierarchy of control systems, but thisproblem is solved on the input side of the hierarchy –when there are several higher order systems.
“The closed-loopmodel of Powers characterizes the actor as an inflexible general-purposedevice”. p 31
No comment necessary –Oh. All right. Can everyone say “reorganization”.
An interesting followup to this catastrophic article (from an HPCTperspective –maybe an accurate report would have allowed the PCT revolution to start in theearly 1980s) is that Carol Fowler was the Editor at JEP who handled the reviewof my 1986 Hierarchical Control paper which was EXTREMELY critical of theFowler/Turvey article. Yet, Ms. Fowler was not only fair in the review process –she was largely responsible for arguing for it’s acceptance (since there weresome disagreements between the reviewers about its merit). So, although CarolFowler happens to be the first author of one of the most notorious anti-PCTarticles ever published, I would have to class her as one of the “good guys” –if not in terms of understanding PCT, certainly in terms of scientificintegrity. I think she must have been a student of Turvey’s at the time thearticle was written –and under his apparently strong “trendy science” spell (I’ve never met him butI hear he talks a pretty convincing line of BS).
And, yes, it’s true –I AM a nice guy; though not nearly as nice as Greg and Ray are for sayingit.
Date: Wed Jan 20, 1993 12:25 pm PST
Subject: Re: feedback too slow
[Avery. Andrews 930121.700]
I don’t think that sending out a list of embarrassing quotes to 50 bignames is the right way to go. After all, it’s not the big names who are theworthwhile targets, but their prospective students, and assorted dissatisfiedmavericks. Priority one is to get a useful document available online forCSGNet, so that people who are intrigued by what we’re up to but not sure thatit isn’t bull can get some easily verified hard evidence that manyestablishment figures are seriously confused. Then perhaps some kind ofarticle somewhere.
Re 6 DoF, has anyone analyzed the DoF of a set of octopus tentacles, Iwonder? (and they can learn to open jars with them …)
Date: Wed Jan 20, 1993 1:52 pm PST
Subject: Schmidt’s Book
[FROM: Dennis Delprato (930120)]
Late last week I promised AVERY ANDREWS I would try to locate R. A.Schmidt’s Motor Control and Learning: A Behavioral Emphasis. I found the 2nded. (1988, Champaign, IL: Human Kenetics Press). Given Greg’s marvelousefforts with the 1st ed., you may not need anything further from Schmidt.However, if you, AVERY, supply me with your regular mail address, I could passon hard copy of any pages I might find that update anything in the 1st ed. Iwant to live up to my promise, but don’t have the flying fingers ofGreg.
Date: Wed Jan 20, 1993 3:53 pm PST
Subject: schmidt not a goodguy
[Avery Andrews 920121.1030]
Here’s some evidence that Schmidt and a number of other people don’tunderstand feedback very well:
Cruse, H, J. Dean, H. Heuer and R.A. Schmidt (1990) `Utilization ofSensory Information for Motor Control’, in O. Neumannn & W. Prinz, Springer-Verlag,pp 43-`79.
This article covers a lot of ground, and makes some proposals that may wellbe right, as far as I know, but the smoking guns of ignorance appear on p. 60,where they are discussing a model in which commands are sent out to effectorson a `feedforward’ basis, with a feedback system supplying corrections (sort oflike what I’ve speculated for driving). The bloopers:
1. “The advantage of an open loop system is that it responds quickly to aninput signal, and cannot become unstable like a closed loop system subject to adelay in the feedback signal.”
AA The first statement being wrong, since a reference level and a currentperception can be compared monosynaptically, which is trivial in the overalltime-budget.The second is true, but irrelevant, since you get the possibility ofinstability once a feedback system is included.
2. “Owing to the operation of the open-loopsystem, the closed-loopsystem only has to deal with small error signals, so the gain (S) can berelatively low.”
AA Wrong because you need a high-gainsystem to compensate for small errors.
There’s lots more here to mull over, but I can’t do it today.
Date: Thu Jan 21, 1993 2:19 pm PST
Subject: devils bibliography -bizzi
[Avery Andrews 930122.0906]
Bizzi’s work seems soundly and closely reasoned, but there’s a bit ofevidence that at least in 1976 he had some strange ideas about how kinestheticfeedback would be expected to work. It appears in
Bizzi, E., A. Polit and P. Morasso (1976) `Mechanisms underlyingachievement of final head position’, J. Neurophys. 39:435-444.
The major point of the paper is that when monkeys orient their headstowards a flash of light, the final head position seems to be determinedwhether feedback is available or cut off (by deafferentation), and that addingeither inertial or constant force loads to the head does not affect the finalposition (once the constant force load is removed).
The strange idea about kinesthetic feedback arises in discussing thepossibility of a mechanism comparing actual with desired vs. actual currenthead position: “the output of this hypothetical comparator might provide asignal leading to the cessation of the ongoing motor pattern.” (rather than thecessation of a signal from the comparator causing a cessation of the motoreffort). This idea is attributed to two earlier papers:
Gibbs, C.B. (1954) `The Continuous Regulation of Skilled Response byKinaesthetic Feedback’, British Journal of Psychology 45:24-39.Eccles, Sabah, Schmidt + Toborikova (1971) `Modes of Operation of theCerebellum in the dynamic loop control of movement,” Brain Research 40:73-80.
Also somewhat suggestive is the conclusion drawn from one of theexperiments. In this experiment intact monkeys orient their heads to a lightflash against a constant force load, which is released soon after the movementcommences. There is no pre-setposition they are trained to achieve, but what happens is the head stops forwhile, and then, when the load is removed, moves a bit further, as if therewere a spring-likeforce attracting it to its final position, so that the initial part of the turnstops when the load balances the spring force.
This is consistent with any number of possibilities (such as a kinestheticreference level with a relatively low gain control system), but the conclusiondrawn is:
“the program for final position was maintained during load application andwas not readjusted by proprioceptive signals acting at segmental orsuprasegmental mental levels” (438).
“proprioceptive signals originating from the moving neck fail to reset thecentral patterns responsible for final position” (442)
True enough, but what is not so clear is why this would be expected tohappen in the first place: what makes this non-reprogrammingsignification? Describing what is basically a fixed setting of some sort as a`program’ also seems a bit odd.
In a later paper,
Bizzi, E., P. Dev, P. Morasso and A. Polit (1978) `The Effect of LoadDisturbances during centrally initiated movements’, J. Neurophys. 41:542-556.
they calculate that the feedback loops (spinal and such higher levelkinesthetic as may be operating) contribute between 10% and 30% of the spring-likerestoring force.
Something that strikes me about the literature I’ve seen so far is thatpeople seem to have the spinal reflex loops reasonably well in hand, but theshow falls apart when it comes to kinesthesis, arguably, I think, due to theprofusion of false ideas about kinesthetic feedback would actually work.
Date: Thu Jan 21, 1993 3:56 pm PST
Subject: kugler et al
It might be worth figuring out where Turvey et. al. get the followingabsurd idea from:
“In control theory, the command-algorithmis separate from the power-fluxthat it modulates; in the neurophysiology of movement, the central nervoussystem is held conceptually separate from the skeletomuscular apparatus thatperforms the movement”
Kugler, Scott, Kelso and Turvey (1980) `On the Concept of CoordinativeStructures as Dissipative Structures’, in Stelmach & Requin (eds) Tutorialsin Motor Behavior, North Holland 3-48.
Absurd because people like Rack, etc. go to great lengths to establish thephysical properties of the power-generators, and their effects on the properties of feedback loop.
Probably out of the same vein as the PCT criticisms that Rick regales uswith.
Date: Fri Jan 22, 1993 11:36 am PST
Subject: kugler et al/robot requirements document
[From Rick Marken (930122.0800)] Avery Andrews says —
> It might be worth figuring out where Turvey et. al. get the followingabsurd idea from:
> “In control theory, the command-algorithmis separate from the power-fluxthat it modulates; in the neurophysiology of movement, the central nervoussystem is held conceptually separate from the skeletomuscular apparatus thatperforms the movement”
This doesn’t seem absurd to me; in fact, it sounds exactly correct. If by”command algorithm” they mean the error signal (or the process that convertsperceptual into error signal) and by “power flux” they mean the forces exertedby the muscles as a result of influence by the error (command) signal then theyare treated separately in control theory –inasmuch as they are treated as separate variables (which they are). The”output” part of the control model says that o = f(e). For systems at thelowest level of the control hierarchy, o could be called “power flux” and ecould be called the “command signal”. They are conceptually separate. Myquestion for Kugler, Kelso and Turvey, then, is SO WHAT? What’s wrong withtreating two separate variables that are functionally related as two separatevariables that are functionally related? The important thing about controltheory is that it also says that e = g(o,p,r,d) –that is, the command signal is (at least in part) a result of the very “powerflux” (o) that it commands. This closed loop relationship MUST be taken intoaccount when analyzing (as you say Rack, etc do) “the physical properties ofthe power-generators, and their effects on the properties of feedback loop”.
So I would say that the above clam is not absurd; it is just irrelevant (atleast, I cannot see it’s relevance to anything); it just seems like part of theconstant desire by conventional psychologists (and roboticists, etc) to saysomething about control models of behavior that might be construed as negative –but is usually wrong or (like the above) a non-sequitur.The goal seems to be to dismiss perceptual control models in order to get onwith the real business of wasting time on “self-regulating”systems (attractors models) and other complex output generation/planningschemes.
Best regards Rick
Date: Fri Jan 22, 1993 11:28 pm PST
Subject: reviews & replies
[Avery Andrews 930123.1825]
I think the reviews & replies proposal is a great opportunity, but italso carries a certain risk of getting clobbered again, since the usual mythsabout feedback are probably all out there, & I don’t think the currentrange of PCT publications deals with them adequately. So I think there’s gotto be some kind of `myths about feedback’ article included to function as apreface. Something along the lines of Bill’s `Objections to PCT posting'(covering feedback too slow and deafferentation), but dressed in full academicbattle gear.
It might be useful to point out that PCT is basically in the same camp asthe dynamics crowd (yes, Kugler, Turvey, etc.) as opposed to `orthodox’computational cog. sci, if there actually is such a thing anymore (in the sensethat I suspect that nobody really sees it as The Truth anymore, although thereare people like me who think that a certain amount of useful work can still bedone from that point of view).
Something else that might be true, & if so, worth pointing out, is thatmisunderstandings about feedback may well have directed research into areassuch as acquired skills that arguably involve a lot of pre-programming,and away from routine but non-stereotypedmanipulative activity.
Date: Sat Jan 23, 1993 12:10 pm PST
Subject: camps, Devil’s Biblio
[From Rick Marken (930123.1100)]
Avery Andrews (930123.1825)
> It might be useful to point out that PCT is basically in the same campas the dynamics crowd (yes, Kugler, Turvey, etc.) as opposed to `orthodox’computational cog. sci, if there actually is such a thing anymore
I liked to be in camps when I was a kid but not so much any more. And eventhen, I would NEVER have wanted to be in the Kugler, Turvey etc camp; I wantedto be in camp to explore, build models and have fun; that’s why I like campCSG.
I understand the desire to find value in the work of those who are not PCTmodelers but who are working on similar problems (like the motor controlproblems of Kugler, Turvey, et at). The downside of trying to control one’sperceptions relative to this desire is beautifully illustrated in today’scontribution to the Devil’s Bibliography. It is a book that Ed Ford suggestedthat I look at:
M.E. Ford and D. H. Ford (1987) Humans as self constructing living systems.New Jersey: Erlbaum.
(No relation to Ed, I presume).
I just got a copy yesterday and it is a goldmine for the Devil’sBibliography. D. Ford seems to be the big theorist here –taking us to new horizons in the development of the control model. D. Ford isactually a modeler in the Carver/Scheier tradition –diagrams are all you get. Also like Carver and Scheier he is a happy camper –he plays contentedly in the camps of Powers, Carver and Scheier, S. Beer(cybernetics guru), Kugler, et al, Pribram, D. McClelland, Bandura, Ashby, etc.Quite eclectic. If you read the overview (chapter one) you might see theproblem with this accepting attitude; in order to have it, you MUST make someglaringly basic mistakes about PCT; M.E. Ford and D.E. Ford make PLENTY.
Here are some examples for the DB.
“Feedback information enables a system to react to events after they haveoccurred. However, for a system to adapt efficiently in a variable environment,it must also be able to anticipate what is likely to happen in the future. Mosthuman behavior is anticipary in nature. Anticipatory actions are accomplishedthrough feedforward processes”. p 9
(Making camp cognition people happy).
“Behavior patterns are a function of the informational-behavioraltransactions with the environment”
(Making camp information processing happy; maybe camp interactionism too?nobody’s been able to figure out what the kids in that camp actually DO though –except be “real scientists”).
“Perception is ‘direct’ and provides accurate information about currentevents within and around a person” p. 22
“Human sensory-perceptualcapabilities are designed to collect information useful for GUIDING PRACTICALACTION in the physically structured and dynamically varying terrestrialenvironment in which humans evolved” p 22 emphasis mine
(camp info processing and camp JJ Gibson)
There are many others in this chapter but I’m getting nauseous.
To be fair, Ford and Ford say many things about control that SOUND close tobeing correct –or that are correct. But this is the problem of dealing with people who onlydeal with control theory as a set of diagrams and phrases –ie.with people who don’t model. It is possible to take almost ANYTHING they say asa correct description of a control system.
I think one way to show that Ford and Ford just don’t really get it (nomatter how often their words might overlap with PCT words) is that they NEVERexplicitly describe the central fact about living control systems –that they control their own PERCEPTION. They talk all around this but manage toavoid this fact about control and, thus, all that it implies about what wewould want to know about the behavior of living control systems –ie. what perceptions they ARE controlling. Ford and Ford (like all those whodon’t get PCT) take behavior at face value –they assume that what THEY PERCEIVE about the organism’s doings is what theorganism IS DOING (controlling). They have quite a way to go; I’m afraid they(like those in the rest of the camps) think they are just about there. Toobad.
Date: Sat Jan 23, 1993 5:32 am PST
Subject: Good, good, and good
From Greg Williams (930123)
>Rick Marken (930122.0800)
> We now apparently have several “real” robotics listening in on thenet. How about asking them to provide a “requirements document” for such arobot.
I second the motion. Chris? John? Other robot researchers? Please!
>Avery Andrews 930123.1825
> So I think there’s got to be some kind of `myths about feedback’article included to function as a preface. Something along the lines of Bill’s`Objections to PCT posting’ (covering feedback too slow and deafferentation),but dressed in full academic battle gear.
Last year, soon after the arm paper was returned, Bill spoke with me aboutworking up a summary of the history of the “feedback too slow” myth forinclusion in an expanded arm paper to be submitted to BIOLOGICAL CYBERNETICS(or some other journal). I haven’t had time to do it, and you seem to be well-immersedin the subject, so why not continue your explorations and write them up? Youare absolutely correct that such an article is needed (whether standalone or aspart of the arm documentation). You’re probably halfway there, at least.Linguistics can wait, right?
As ever, Greg
Date: Sat Jan 23, 1993 2:57 pm PST
Subject: throwing stones from glass houses
[Avery Andrews 930123.930]
One problem with my writing the contra-feedbackmyths article is that in many respects I’m in the same situation as the peopleI’d be criticizing. E.g. in my post about the Schmidt article I said that youneeded a high-gainsystem to correct small errors, but while walking the dog this morning Irealized that the magnitude of the disturbing forces is what matters, not theabsolute size of the errors.
One of the obstacles to PCT understanding is perhaps that it is aprofoundly Newtonian theory, while most people’s world views are pre-Newtonian,and the Newtonian picture is *hard* to acquire.
Date: Sat Jan 23, 1993 4:06 pm PST
[Avery.Andrews 920123.1050] Rick Marken (930123.1100)
> “Feedback information enables a system to react to events after theyhave occurred.
So here we have the event blunder. I’d say something along the lines thatevents are useful for certain purposes, but we can’t assess their overall roleadequately without properly understanding the continuous aspect ofthings.
> However, for a system to adapt efficiently in a variable environment,it must also be able to anticipate what is likely to happen in the future. Mosthuman behavior is anticipary in nature.
I’d be happy with this if they replaced `most’ with `much’.
> “Behavior patterns are a function of the informational-behavioraltransactions with the environment”
Gobbledegook (we agree here, I think)
> “Human sensory-perceptualcapabilities are designed to collect information useful for GUIDING PRACTICALACTION in the physically structured and dynamically varying terrestrialenvironment in which humans evolved” p 22 emphasis mine
Hmm. This sounds fine to me, at least if `are designed to collect’ isreplaced by `have been selected for success in collecting’.
Date: Sat Jan 23, 1993 6:50 pm PST
Subject: myths article
[Avery Andrews 930123.1130]
Wobbly as my grasp & background may be, I do have some ideas about howa feedback myths article might be organized, which go as follows:
1. Some examples of egregious blunders (to get people to see quickly thatthere is a problem).
2. The major myths & how they started.
3. Some bad effects on current research.
4. Some puzzles & anomalies that might be cleared up by the BCPapproach.
Date: Sat Jan 23, 1993 6:51 pm PST
Subject: devil’s bib entry
[Avery Andrews 930123.1350]
Schmidt, R.A. (1980) `On the Theoretical Status of Time in Motor-Program Representations’, in Stelmach & Requin, _Tutorials in MotorBehavior_, North-Holland,pp. 145-166.
Discusses experiments in which levers are moved quickly, in opposition tovarious sorts of disturbing forces. The results suggest that any positionalcontrol has fairly low gain, but the article suggests unawareness of the roleof gain in the functioning of feedback systems:
“A feedback view would, of course, predict that the limb would reach thetarget (against a constant-forcedisturbance) …” (p. 159).
“The effects of added or subtracted spring tension on the movementendpoint support the mass-springnotion, and provide additional evidence against the idea that the terminalposition is achieved by some sort of feedback process.” (p. 160).
Also betrays possible confusion of continuous control with proprioceptivelymediated response chaining:
“This argument was strengthened by human evidence that the processing ofinformation leading to a new movement was slow, requiring 150 to 200 msec forthe new action to begin. This kind of feedback processing, if it were to beemployed in the ongoing control of a rapid motor act like throwing, would betoo slow to be effective until that act is completed.” (p. 148)
I am frankly not at all sure what’s being addressed here: throwing is a*continuous* act, not one that is cleanly segmentable into discrete subacts, soit’s not clear how response chaining would work; furthermore initiating avoluntary act is presumably different from ongoing modification of a `program'(and, Gary Cziko has a demo to the effect that accurate throwing is possible inthe face of disturbing forces -a quantitative study would seem called for).
An important point is that PCT does not challenge the existence of CPGs,rather, it simply claims that they will normally produce reference levels forperceptions (and will therefore in general be able to produces error signalsand drive behavior when the afferent pathways or cut, although considerableretuning will be necessary to get passably effective behavior).
On a more positive note, the article contains a lot of interestingmaterial, and does refute the impulse-timingview of movement control, which is more counter-PCTthan the mass-springview, and is open to the idea that afferent information modifies the outputs ofCentral Pattern Generators (p. 147).
It is proposed that programs have a variety of parameters, some of whichseem more plausible than others. E.g. Movement time seems plausible (thepattern runs faster or slower), while Force and Muscle Selection seem verydubious. I suspect that Gary Cziko’s throwing demo can be turned into a totalrefutation of the Force parameter (if you can throw accurately against variabledisturbing forces, you can’t do it with a preset Force parameter).
There is perhaps an avenue of empirical investigation into Muscle Selectionas well. People appear to have a fixed handwriting style that is invariantover a substantial size range, from blackboard writing done with arm muscles,to ordinary writing done with fingers. This suggests that a size-scalableperceptual target is involved, perhaps involving kinesthetic effortperceptions, etc. If so, then *deafferented* people would not be expected tohave a size-scalablehandwriting style, at least when their eyes were closed. Writing on theblackboard with your eyes closed and rubber bands attached to your arms mightreveal things as well. ed
Date: Sat Jan 23, 1993 9:38 am PST
Subject: Re: Good, good, and good
I don’t think I’m halfway there to a feedback too slow paper at all. Theliterature is huge, I’ve looked at only a tiny portion of it, & there arepeople out there in CSGNet with much better backgrounds for interpreting itthan I have. Also, linguistics can’t wait much longer, since teaching startsup again in a month down here. So I can do some of it, but not all ofit.
Date: Sun Jan 24, 1993 8:40 pm PST
Subject: Re: camps, Reviews
[Avery Andrews 920125.1529] (Rick Marken (930124.1000)
> If anything is “guiding practical action” in a control loop, it is thenet effect of the disturbance on the controlled perceptual variable.
Well, that’s pretty much what I interpreted them as saying, but youactually say it, rather than just allowing me to read you that way. Thevagueness and ambiguity of a lot of this literature is certainly a majorproblem with it, which deserves a lot of attention.
Another likely devil’s bib entry is:
Adams, J.A. (1971) `A Closed Loop Theory of Motor Control’, Journal ofMotor Behavior 3:111-149.
A survey of recent theories of motor behavior that I’ve been looking at,and may report on shortly, cites this as the origin of the (supposedly defunct)`closed loop’ paradigm.
Date: Sun Jan 24, 1993 9:14 pm PST
[Avery Andrews 930125.1610]
The bad guys say:
Perception guides action
What Rick Marken says may (I hope) be paraphrased as:
Perception of the net effects of disturbances on a controlled variableguides action
Which is clearly more specific, & therefore more useful.
Date: Mon Jan 25, 1993 12:12 am PST
Subject: Re: myths article
[From Oded Maler (930125)]
Re: Feedback too slow
I posted this inquiry to a relatively-prominentresearcher in motor control, and the following answer contains some referencesthat might be interesting.
> Secondly with regard to your question-there is out from December the last issue on Current Contents in Neurobiology-motorcontrol with reviews about role of feedback (either by Hasan or Laquaniti Idont remember) A couple of years ago there was also a review by Hasan andStewart in a book in the series Excersise and Sport Science Reviews (1988 or1986 or 1987). It is generally a matter of debate-a lot of things can be done without feedback. Actually there is also a paper byGhez in the Cold Spring Harbor Symposium on the Brain (1990).
Hope this helps –Oded
Date: Mon Jan 25, 1993 4:45 pm PST
Subject: Re: More Devil’s Bibliography
> I think she must have been a student of Turvey’s at the time thearticle was written –and under his apparently strong “trendy science” spell (I’ve never met him butI hear he talks a pretty convincing line of BS).
I believe that they’ve married since then. I’ve also experienced a
poor “response rate” from her/them (i.e., answering letters).
Date: Tue Jan 26, 1993 11:28 pm PST
Subject: Abs & Winstein (feedback too slow)
[Avery Andrews 930127.1710] Rick Marken (930112.0800)
> If you want to read more amusing statements about feedback being “tooslow” made by authoritative leaders in the study of human movement control ,try the article by Abbs and Winstein in M. Jeannerod (ED) Attention andPerformance XIII, Hilldale, Erlbaum, 1990
I had a look at it, and didn’t find the discussion about speed silly atall. They debunked the 200ms myth, showed that feedback thru the oral trackcould occur in as little as 12 msec, noted that feedback from the distal armswas faster than the proximal arms, maybe in the 30 msec range, etc. It isn’t myfield, but I didn’t see any figures that were at variance with common sense,& the general trend of the literature.
What I did find intriguing was the following assertion:
“Technically, a feedback system is one in which an error signal directlydrives a corrective adjustment >at the site where the error isintroduced<” (p. 366, my emphasis)
This was supposed to entail that compensatory lip adjustment couldn’t befeedback (this is when someone is trying to make, say, a /w/, which requiresthe lips to come close together, and one lip is disturbed, & the other goesfurther to make up for it).
Does anyone know any basis for this `technical’ restriction on the scope offeedback? It seems patently wrong, even on the basis of the classic examplesof thermostat and ships rudder, where the error might be introduced in fronthall, when Fred leaves the front door open for a while, and the compensatoryadjustment is made in the basement by the furnace starting.
I didn’t have the time to read the whole article carefully, but I didnotice that the authors are caught up in the stampede of enthusiasm for `motorprograms’, which still appear to me to be a rather vague and woolly conceptthat we might hope to get somewhere by replacing with the ECS.
Date: Thu Jan 28, 1993 3:43 pm PST
Subject: feedback too slow (summary)
Here’s what I get for the `feedback too slow’ issue (for humans -insects, etc. is a whole nother ballgame). It’s not really an article-weightsubject, since the standard literature actually seems to have sorted itself outpretty well. To talk about feedback w.r.t. high-speedmovements (the jab of Muhammad Ali, professional piano trills) you clearly needto really know physiology and the mathematics of control theory, while forordinary-speedactions it’s entirely clear that feedback is fast enough. One way in whichthis could be improved is to find more standard references for the basic facts,so that anybody could find one in a library near them. But the ones given aresufficiently authoritative.
The next topic I want to look at is PCT versus Central Pattern Generators:the `closed loop’ theories of Jack Adams (the very one who wandered into Bill’sclass, almost certainly) got clobbered when people became convinced of theexistence of CPGs, but this is a non-issuefor PCT, since there’s no reason why there can’t be CPG’s whose output istreated as a reference level. But this may take longer -I’ve essentially spent the last month full time on PCT, but I can’t keep ondoing this.
People who make generic claims that feedback is too slow are probably (a)thinking of highly practiced rapid movements (b) thinking of the 200-150ms reaction time for visually presented stimuli. But more recent work showsthat ongoing actions can be modified by sensory information much more quicklythan that, as reviewed by for example by Schmidt (1982:216-220),Schmidt (1988:164-177),Abbs and Winstein (1990:631-635). (Note the more-than-doublingof the length of the section of Schmidt devoted to this issue. The discussionof the `wineglass effect’ in the 1988 book is particularly interesting from aPCT perspective (Johanssen amd Westling 1984, Westling and Johanssen1984.)
There are two kinds of effects that have been studied for various kinds ofmuscles, the `short latency’ spinal reflexes, with latencies ranging from 30ms(upper arms), 15 ms (fingers) to 6ms (orofacial). These have relatively lowgain, and don’t do much to resist serious applied disturbances. Then there arethe `long latency’ reflexes involving the brainstem and cerebellum, whoselatencies range from 70ms (upper arms) to 44-55ms(fingers) to 20-30ms(orofacial).
From this it seems clear that feedback will not be very useful incontrolling certain high-speedmovements, such as the left jab of Muhammad Ali, clocked at 40ms. On the otherhand it might be possible to trigger it through a kinesthetic feedback loop:set a reference level for a distant position of the fist, then step change fora closer one (deceleration + return), with the timing pre-arrangedso that the deceleration is mostly achieved via elastic restoring forces fromthe opponent’s body. But the feedback would not be helping to solve anyproblems, but would just be creating them (when the grossly out of date sensoryinformation about the movement comes back up the line). Polit and Bizzi (1979)also have some evidence that highly practiced pointing gestures may be acquiredas patterns of alpha-gammaactivation, since these gestures can be produced accurately with minimal changeonly 2-4days after deafferentation (much shorter than the usual recovery period), aslong as the starting posture is the same as when the gestures were learned, andno disturbances are applied.
Piano trills, on the other hand, don’t offer as severe a problem as onemight expect. If a trill contains 16 notes per second, then the two fingerswill be cycling at 8 cps, giving a period of 125ms, which should be manageable(barely) by a loop with a 55ms lag (and maybe professional pianists can reducethese lags somewhat). [It would be good for an actual control theorist to checkthis -I’m just repeating stuff I’ve read without understanding the mathbehind.]
In general then, the question of whether feedback is too slow has to beasked for specific kinds of movements involving specific muscles, and forordinary manipulations of objects the answer is clearly that it is not,although certain fast & practiced movements are obviouslyexceptions.
One issue that perhaps deserves mention is the speed of feedback in theinitiation of `responses’. Feedback control of the initiation of a responsecan be thought of as modulating the production of the response w.r.t the extentto which its intended effects already exist. But most significant responses(such as the withdrawal response from heat) are already heavily modulated byvarious kinds of sensory information, so adding in information about the extentto which the effect of the response already exists is not going to addsignificantly to the time budget. Camhi’s overall assessment is (p.111):
“Indeed, there is increasing reason to believe that reafferent feedbackplays a significant role in the control of most or all categories of behavior.Neither the speed of the movements nor their directional or other propertiesshould be used as _a priori_ argument against their involvement of suchfeedback.”
The serious speed-of-feedbackissues are (a) stability when the frequency of variations in the referencelevel is too fast w.r.t. the time-delaysn the system (b) whether feedback can be helpful in adjusting the ongoingresponse to current conditions. These problems need to be looked at on a case-to-casebasis, with full knowledge of the relevant technical details. In many cases,it seems to me that the results look pretty good for PCT. For example, Camhi(1985) argues for the plausibility of closed-loopcontrol of the rate of turn in the `escape-reflex’of the cockroach.
A final contaminating factor deserves mention: in Robotics, it seems tonormally be assumed that paths and trajectories should be plotted in advance,which involves considerable computational expense (as discussed briefly bySchmidt, for example). This may well be appropriate for robots, since it isperhaps not a good idea to have these expensive, dangerous and very stupidmachines deciding what they are going to do on a moment-to-momentbasis. Furthermore, in order to make money for their owners, robots have toreplace highly practiced assembly-lineworkers, so that you need something to play the role of practice, which willeither be actual practice (adaptive control) or planning. But considerationsthat are important for Robotics are not necessarily the main priorities fornormal human movement.
Abbs. J.H. and C.J.Winstein (1990 `Functional Contributions of Rapid andAutomatic Sensory-basedAdjustments to Motor Output’, in Jeannerod, M. (ed) _Attention and PerformanceXIII_, 627-652.
Camhi, J.M. (1985) `Feedback Control of an Escape Behavior’, in Barnes,W.I. & P. Gladden, eds. (1985) _Feedback and Motor Control in Invertebratesand Vertebrates_, Croom Helm 93-111.
Johnanssen, R.S. and G. Westling (1984) `Roles of Glabrous Skin Receptorsand Sensorimotor Memory in Automatic Control of Precision Grip When LiftingRougher or more Slippery Objects’, Experimental Brain Research 56:560-564.
Polit, A. and E. Bizzi (1979) `Characteristics of motor programs underlyingarm movements in monkeys’ J. Neurophys. 42:183-194.
Schmidt, R.A. (1982) _Motor Control and Learning: A Behavioral Emphasis_,Human Kinetics Publishers, Inc.
—(1988) _Motor Control and Learning: A Behavioral Emphasis, 2nd ed. HumanKinetics Publishers, Ic. L
Westling, G. and R.S. Johnassen (1984) `Factors Influencing the ForceControl during Precision Grip’, Experimental Brain Research 53:227-284.
Date: Thu Jan 28, 1993 10:25 pm PST
Subject: Re: feedback too slow (summary)
[From Rick Marken (930128.2200)] Avery.Andrews (930129.1000)
> People who make generic claims that feedback is too slow are probably(a) thinking of highly practiced rapid movements (b) thinking of the 200-150ms reaction time for visually presented stimuli.
Before starting the article (or whatever it is) here, why not spend amoment explaining what the hell people might mean by the idea that “feedback istoo slow”. I think the whole concept is ridiculous because it is based on asequential state, cause effect concept of a control loop –one which leaves out time (and, hence, a large hunk of reality). How canfeedback be “too slow” in a closed loop where feedback is present (as aperception) at the same time that the cause of that feedback (error) ispresent. There are phase relationships between the continuous variables in theloop –theresult of transport lags and slowing factors around the loop. Lags that are toolong or slowing factors that are too great can create instabilities in the loop(they can also create stabilities). But it would be necessary to know what ismeant by “feedback too slow” and then to test it in a model before one couldsay whether “feedback too slow” (what ever that is) would be a problem. Infact, without a definition of what “feedback too slow” means in term of closedloop control, it is difficult to know what is being measured in these reactiontime experiments. If you apply an impulse disturbance to a controlled variable,there will be a change in the output variable; what is the “reaction time”here? The time to maximum output, the time until this output has some othereffect (like pressing a button), the time until the derivative of the output ismaximum, minimum? And when you decide what reaction time is, has it beenmeasured in the same way by all these experimenters? And if it has, what is thereason for this reaction time -is it a transport lag, slowing factor, some of both? Without a model, how canthey even tell what the reason for the reaction time might be?
The point is that most of the data presented in these papers is probablyuseless because it is not collected in the context of a working control model.There are lags and slowing factors in control systems so people are bound tofind response latencies when they apply sudden disturbances to things peopleare (or are expected to be) controlling; the observed results in these studies(16 ms, 150 ms) may look very scientific and all but they are almost certainlyuseless for modelling behavior. And there is no question that measures of”response latency” that are collected in this way (with no understanding of thebehavior of closed loop systems) say nothing about what variables people canand cannot control. Using these reaction times as a basis for showing thelimitations of feedback control is just silly –and an impediment to real research on control. This is what the article shouldbe about.
Date: Thu Jan 28, 1993 11:27 pm PST
Subject: feedback too slow (summary)
[Avery Andrews 930129.1823] Rick Marken (930128.2200)
> Before starting the article (or whatever it is) here, why not spend amoment explaining what the hell people might mean by the idea that “feedback istoo slow”.
Because I suspect that most of the people who repeat this slogan don’t meananything by it, but are just repeating a formula they’ve picked up in order toavoid doing some work. It’s useless to try to analyze every single waysomebody might get into this frame of mind -what is useful is a bit of prose showing that even from the conventionalviewpoint you can’t just assert this.
> And there is no question that measures of “response latency” that arecollected in this way (with no understanding of the behavior of closed loopsystems) say nothing about what variables people can and cannot control. Usingthese reaction times as a basis for showing the limitations of feedback controlis just silly –and an impediment to real research on control. This is what the article shouldbe about.
Well, I’m just not game to go around claiming that Houk & Rymer, orP.M.H. Rack, or Abbs & Winstein don’t know what they’re doing. If you are,go ahead.
Date: Thu Jan 28, 1993 11:30 pm PST
Subject: Adams (1971)
I started looking into Adams 1971, & it looks like it will be a realmotherlode for the devils bibliography. Exhibit 1:
“There is a reference that specifies the desired value for the system, andthe output of the system is fed back and compared to the reference for error-detection,and, if necessary, corrected. The automatic home furnace is a common example.The thermostat setting is the desired value, and the heat output of the furnaceis fed back and compared against this reference” (Adams, J.A. `Closed LoopTheory of Motor Learning’, JMN,m3:11-150;p. 116
Here the blunder is that what the thermostat is measuring is the actualoutput of the furnace, as opposed to the result of the furnace output & allother influences on the air temperature in the immediate vicinity of thethermometer. This mistake is probably due to the typical Wiener-styleand chemical engineering diagrams of feedback systems (like the ones at thebeginning of Houk and Rymer 1981), where you have a box labelled `controlledsystem’ with the comparator, effectors, etc. outside of this box (maybe this iswhere Kugler, Turvey et.al. get their strange ideas about control theoryfrom).
This blunder cross-fertilizeswith another, the apparently classic distinction between exteroception andproprioception, which figures in a quote on the next page:
“For James, feedback acts as stimuli, and has no more status than anexteroreceptive stimulus which starts the sequence, like a light on adisplay.”
Of course, James was just talking about response-chaining,but he seems to have been closer to the right idea nonetheless.
So maybe a point to emphasize when talking to psychologists is that in PCTthere is no `proprioception’ or `exteroception’, but just `perception’ (thisactually is rather Einsteinian -maybe I’m getting the point of what Martin was saying a few days ago). Thispoint can be enhanced by observing that Schmidt (1982, 1988) discussing someproblems with these notions of perception, cites with approval a proposal tointroduce a third, blended term, `exproprioception’ (for `movements of our bodyin relation to the environment’).
Date: Fri Jan 29, 1993 8:16 am PST
Subject: Slow feedback
[From Bill Powers (930128.2200)] Avery Andrews (930129.1000) —
You’ve made a start on a formidable paper. I hope that others will joinwith you in bringing it to completion. I think it will be known in the futureas a fundamental work in the field of PCT.
I would like to clear up one point that is still unclear, which is controlof fast movements, or rather the obvious lack of it:
> From this it seems clear that feedback will not be very useful incontrolling certain high-speedmovements, such as the left jab of Muhammad Ali, clocked at 40ms.
Imagine a control system that has a time-constantof 40 milliseconds (we don’t actually have to imagine it; the Little Man modelsit). This control system can control the position of an arm, so that one ormore perceived joint angles follows a smoothly varying reference signal, movingthe arm under complete control at all times.
But now take this very same control system, and instead of giving it areference signal that passes smoothly from one magnitude to another, make thereference signal jump instantaneously –in, say ten milliseconds –from a frequency of 200 impulses per second to a frequency of 500 impulses persecond, with no passage through intermediate frequencies.
The arm will move from the initial position corresponding to 200 impulsesper second to the final position corresponding to 500 impulses per second. Ifthe control system is critically damped, it will do so in a single swift movethat goes to the new position in an exponential approach. With a time constantof 40 milliseconds, it will go 63% of the way to the new position in 40milliseconds, 86% of the way in 80 milliseconds, and 95% of the way in 120milliseconds. This represents the fastest speed at which the control system cancorrect a suddenly-appearingerror of any magnitude, large or small.
During this maximum-speedmovement, there is naturally no control. The control system is alreadyproducing the largest output it can consistent with stopping in the newposition. Any disturbance that came and went during the approach would simplycause a deviation from the nominal path. So, paradoxically it may seem, thecontrol system’s control action is uncontrolled after a step-changein the reference signal –although the final position is as controlled as ever.
This is simply the nature of control. As the speed of movement increases,due to more and more abrupt changes in the reference signal, the error signalbecomes larger and larger (producing faster and faster movement) until thechange becomes a true step-change, at which point the movement will reach its maximum speed. Theresistance to disturbances that appear during the movement will be, at lowspeeds, essentially the same as for static reference signals. As the rate ofchange of the reference signal increases, the resistance to disturbance duringthe transition becomes smaller and smaller, until in the limit it vanishes.There is therefore no inconsistency between saying that slow movements arecontrolled while the fastest ones are not. There is no need to posit oneorganization for slow movements and another for fast ones. The same controlsystem, with the same parameters, explains the behavior we see over the wholerange.
There is another side to this story. One common idea about the hierarchy ofcontrol is that within it, goals are set and then the control systems alterperceptions to match them. But this appearance, I think, is misleading. On thetime scale appropriate to any level of control, behavior is not a process oferror correction. On that time scale, errors never become large. Instead, asreference signals vary, perceptions simply follow them. The changes we seereflect changes in reference signals, not the process of error correction. Onthe appropriate time scale, which has been called the “specious present,” errorcorrection takes no time.
Best to all, Bill P.
Date: Fri Jan 29, 1993 10:25 am PST
Subject: What is feed-backtoo slow for?
[From Oded Maler (930129)] Rick Marken (930128.2200) Avery.Andrews(930129.1000)
>> People who make generic claims that feedback is too slow areprobably (a) thinking of highly practiced rapid movements (b) thinking of the200-150ms reaction time for visually presented stimuli.
> Before starting the article (or whatever it is) here, why not spend amoment explaining what the hell people might mean by the idea that “feedback istoo slow”. I think the whole concept is ridiculous because it is based on asequential state, cause effect concept of a control loop –one which leaves out time (and, hence, a large hunk of reality). How canfeedback be “too slow” in a closed loop where feedback is present (as aperception) at the same time that the cause of that feedback (error) ispresent. There are phase relationships between the continuous variables in theloop –theresult of transport lags and slowing factors around the loop. Lags that are toolong or slowing factors that are too great can create instabilities in the loop(they can also create stabilities). But it would be necessary to know what ismeant by “feedback too slow” and then to test it in a model before one couldsay whether “feedback too slow” (what ever that is) would be a problem. Infact, without a definition of what “feedback too slow” means in term of closedloop control, it is difficult to know what * is being measured in thesereaction time experiments. If you ply an impulse disturbance to a controlledvariable, there will be a change in the output variable; what is the “reactiontime” here? The time to maximum output, the time until this output has someother effect (like pressing a button), the time until the derivative of theoutput is maximum, minimum? And when you decide what reaction time is, has itbeen measured in the same way by all these experimenters? And if it has, whatis the reason for this reaction time -is it a transport lag, slowing factor, some of both? Without a model, how canthey even tell what the reason for the reaction time might be? etc.
I think this exchange clarifies some important points and shows which partsof the elephant’s body (to use the by-now-classicalmetaphor) are observed by “blind” non-PCTersand which by “blind” PCTer.
The answer to the question “what is feed-backslow for” you must invoke some “objective” performance criterion independent ofthe internal perceptual coordinates of the acting individual. You must assumethat “playing a piano trill correctly at some speed” or “knocking out a boxingchampion” has some more or less agreed-uponmeaning. Then you can build a mathematical model of that act and the componentinvolved (muscles, nerves and their reaction time) and show WITHOUT USING THECONTROL MODEL that it is impossible for information to travel and affect themuscle at the time scale between the initiation of the action and its outcome.The nature of these impossibility/lower-boundarguments is that they consider ideal situations and thus apply as well to the”correct situation”. By showing that an ideal pianist or boxer with ideal”objective sensors” and “objective effectors” cannot achieve something becauseof timing constraints you show a-forteriori(?) the a realistic (i.e. PCT-based)pianist/boxer with the same timing constraints cannot do it either. It is thesame like proving, based upon bio-chemicaland physical reasoning that no human is capable of, say jumping above 10m. Thisargument is true, regardless of whether he is commanding the muscles viahierarchical servoing, inverse-dynamicscalculations or coin tossing. (I think that some of Martin’s attempts explaininformation-theoreticconstraints were along a similar line, and maybe he was right in the statementhe made back then concerning the “real” understanding of PCT 🙂
The emphasis of the important PCT insight that within the individual “it’sall perception” in contrast to the naive objectivism of, say, cognitivepsychology, should not be exaggated into a solipsist neglect of the externalenvironment. The question of how and under what conditions people can achievecertain “objective” performance, in other words, what guarantees that a systemorganized in a certain way survives (“objectively”) in a given environment,deserves more attention and better answers than “otherwise, reorganization willcontinue”.
> The point is that most of the data presented in these papers isprobably useless because it is not collected in the context of a workingcontrol model. There are lags and slowing factors in control systems so peopleare bound to find response latencies when they apply sudden disturbances tothings people are (or are expected to be) controlling; the observed results inthese studies (16 ms, 150 ms) may look very scientific and all but they arealmost certainly useless for modelling behavior. And there is no question thatmeasures of “response latency” that are collected in this way (with nounderstanding of the behavior of closed loop systems) say nothing about whatvariables people can and cannot control. Using these reaction times as a basisfor showing the limitations of feedback control is just silly –and an impediment to real research on control.
Maybe Rick is right about that, but still some neurophysiological evidenceabout basic properties of nerves and muscles can replace this data and provethe uselessness of feedback for certain kinds of actions.
Date: Fri Jan 29, 1993 10:27 am PST
Subject: slow feedback
[Avery Andrews 930130.0420] Bill Powers (930128.2200)
I think I’ve taken the too-slowissue as far as I can without learning a lot more fundamental stuff. & mysense of the subject is that to go much further you’d probably have to do realresearch on specific types of movements. But if the dumb objections thatpeople have to PCT can be sorted out, I think the trills & jabs will prettymuch take care of themselves. & I hope my little piece is sufficient todeal with the ignorant forms of the too slow argument. My perception is thatthe next issues to look at are (a) pattern-generators(b) the idea that the output of the effectors is what is controlled (as in theJack Adams quote, and Abbs & Winstein’s `technical’ definition of feedback.This is sort of like what you call the `objectification blunder’ in QAPR, but Iseem to want to call it the `output blunder’.
I think I’ll have to pass on saying more about language for the moment -it really is a much more difficult collection of issues.
Date: Fri Jan 29, 1993 3:01 pm PST
Subject: FB 2 slow; devil’s bib
[From Bill Powers (930129.1300)] Rick Marken, Avery Andrews(930129)
Response latency is measured in a lot of ways, as Rick says. Most of themleave you with little idea of what the actual transport lag through the nervoussystem is. I saw something recently in which subjects indicated a response bysaying “HA!” into a microphone. Apparently the experimenter thought this wassuch a simple thing to say that saying it required no time. I suspect that anylinguist would be laughing by now, thinking of what a sonogram of “Ha!” lookslike. Just imagine the diaphragm beginning to tense, the pressure building upand starting to leak through the throat, the hiss, and finally the “Ah” soundbuilding up to a measurable level. By the time that sound starts, the momentthat the nervous-systemoutput actually began is fading into the distant past.
Or consider indicating a response by pressing down on a key, or releasing akey. Depending on the relative sensitivity of the shoulder-musclesystems, the biceps and triceps, and the forearm muscles that operate thefingers (not to mention the strength of the spring under the key), the firsttendency of the finger on the key to move might be either in the right or thewrong direction. If the biceps/triceps respond the most, initially, thedynamics of the arm will make the finger press down harder instead ofreleasing, or rise further off the key instead of depressing. By adjustingparameters you can make the Little Man model do these things quite clearly, tovarying degrees. So an unknown part of the “response latency” consists of thedynamics of arm segment movements under angular acceleration.
And of course as Rick said, where do you place the threshold for detectingpresence of absence of a response? At the 10% point? 50%? 90%? Just above thenoise level? The inflection point in the movement? The point of maximumvelocity?
When Bob Clark and I measured reaction times for mechanical disturbances ofhuman arms, we used an electromyograph so we could pick up the moment whennerve impulses reached the muscles; we ignored the actual mechanical movementsoccurring after that. To my mind that is the only meaningful way to talk abouta reaction time.
Heck, I’ve seen experiments with rats in mazes in which “response latency”was defined as the time it took the rat to get from the photocell at theentrance to the maze to the photocell at the entrance to the box where thecheese was.
Avery has it right: “Feedback is too slow” is a slogan, repeated becausesomeone else said it in an authoritative manner.
Rick’s question is highly germane: too slow for what? If feedback is tooslow for control within 40 milliseconds, it’s too slow, even if it’s present.This is true in a model and also in the real system. The feedback pathwaysaren’t surgically removed just before a 40-milliseconddisturbance occurs. They just don’t do much good. Look at the patellar reflex(knee-jerk).The rubber hammer puts in an impulse stretch of the tendon. The control systemtries to correct the error, but it’s much too late; the disturbance is gone. Sothe leg kicks upward, trying to oppose a disturbance that’s not there any more.Feedback was too slow. So what? It was still a control system.
In the discussions of fast and slow movements, there seems to be an ideafloating around that the nature of the physical system changes depending on thespeed of the movement. This can’t be true. Either you have a system hooked upas a control system, or you don’t. If the control system isn’t present for fastmovements, it isn’t present for slow movements, either. If a feedbackconnection is present for slow movements, it’s still there during fastmovements even if it’s responding too slowly to do any good.
It doesn’t matter whether you practice a movement a lot or are doing it forthe first time. The parameters of control may slowly improve with practice, butyou can never get to the point where you can dispense with the feedback. Infact, the better you get, the more important the feedback becomes in assuringan accurate action despite the speed. No matter how good the control gets, onthe other hand, you can always present the system with a step-change in reference signal, or a disturbance in the form of a brief enoughimpulse, so the system is asked to perform beyond its capacities. In that casethere’s no control during the transition. But nothing about the control systemhas physically or functionally changed. If you took away the feedback, themovement wouldn’t end up in the right place.
I hate to say it, but such statements about fast and slow or practice andunpracticed movements simply show the lack of a working model in thebackground.
Oded Maler (930129) —
> The answer to the question “what is feed-backslow for” you must invoke some “objective” performance criterion independent ofthe internal perceptual coordinates of the acting individual.
Feedback is too slow to permit a person to catch a passing bullet. So what?Arbitrary objective criteria don’t have anything to do with the interior designof the behaving system, which behaves to achieve its own aims, not those of anexternal observer. The control systems in the human body are exactly fastenough to permit the kind of behavior that human beings are able to accomplish.They are not fast enough to accomplish ends that human beings can’t accomplish,even though an artificial control system with faster feedback might be able todo them.
> You must assume that “playing a piano trill correctly at some speed”or “knocking out a boxing champion” has some more or less agreed-uponmeaning.
This won’t help to answer the question of whether feedback is too slow toexplain any particular instance of behavior. I certainly can’t play a pianotrill “correctly” in terms of objective criteria that might apply to my friendSam Randlett, who teaches concert pianists. Yet my feedback is exactly fastenough to account for the speed with which I CAN play a trill (Sam would fallasleep waiting for the next note). I can’t knock out a boxing champion, or adrunk paraplegic, yet my feedback control systems are just exactly fast enoughto make my fist move as fast as it actually moves.
> Then you can build a mathematical model of that act and the componentinvolved (muscles, nerves and their reaction time) and show WITHOUT USING THECONTROL MODEL that it is impossible for information to travel and affect themuscle at the time scale between the initiation of the action and itsoutcome.
Again this misses the point. With the Little Man arm model, I can find themaximum speed of movement at which it is impossible for feedback signals tomake any change in the movement between its initiation and its outcome. Thatspeed is one in which a movement occurs in response to a step-changein the reference signal, and its time constant is about 40 milliseconds, justlike Mohammed Ali. The initiation of the act consists of an instantaneouschange in the reference signal. The control system begins with maximum possibleerror, which then reduces asymptotically to zero with a 40-millisecondtime constant. This is simply the fastest speed at which the error can bereduced.
The determination of the delays is quite independent of the model; it’s dueto neural transport lags and to the leaky-integrator form of the muscle response. The feedback model incorporates thesefactors, and when the parameters are adjusted for the nicest and fastest errorcorrection possible under those conditions, you get a 40-millisecondtime constant of error correction. The feedback model is acting as fast asphysically possible.
If you didn’t have the feedback present, the model wouldn’t be able toexecute the same movement anywhere nearly as fast. The feedback model starts,immediately after the step-changein reference signal, with an error signal that would move the arm somethinglike 10 times as far as the actual distance to the reference or intended end-point.This produces a tremendous acceleration, on the order of 20 gravities in linearterms. But as soon as the arm begins to move, the error signal begins to drop(about 9 milliseconds later, actually). Now the effective target location isnot so far beyond the intended end-point.This process continues, the effective end-pointcoming back inward toward the intended end point while the arm moves outwardtoward the intended end-point.Before the arm reaches the endpoint, the rate feedback actually moves theeffective target point to the negative side of the intended end-point,decelerating the arm. All this happens automatically with no particularcomputational difficulties, and the arm comes to rest at exactly the rightposition, 100 or 150 milliseconds after the initiation.
If you wanted an open=loop system to bring an arm to an endpoint after astep-changein the initiating signal, you would have to use a much smaller signal, onetenth as large, to avoid overshooting the intended end-point.As a result, the initial acceleration would be much less, and the movementwould take far longer.
Actually, the problems would be much worse than that. You couldn’t actuallyuse a step-changein the driving signal, because without the feedback you’d have a mass on aspring with very inadequate damping. You would need a full blown motor programthat would apply a complex waveform to the muscle, to prevent oscillations. Ofcourse you could then use a larger driving signal, and in fact supply the samedriving signal that would be observed during the operation of the feedbacksystem. You could then achieve equal speed –but at what cost!
> By showing that an ideal pianist or boxer with ideal “objectivesensors” and “objective effectors” cannot achieve something because of timingconstraints you show a-forteriori(?) the a realistic (i.e. PCT-based)pianist/boxer with the same timing constraints cannot do it either.
This is more like my point. In fact, real boxer/pianists come very close toachieving the theoretical limits of performance. Models, of course, can alwayswork better than real people, because we can give them more favorableproperties. But if we match masses, delays, and time-constants,we can then adjust the model to reproduce the human behavior reasonably well.The Little Man, given very strong muscles, can in fact move the fingertip fromone point to another with a 40-millisecondtime constant, about the same as Ali or Joe Louis. But not a lot faster, giventhe delays and time-constants.Trying to push for still faster movements takes the system to the edge ofinstability. I presume that’s why the real system doesn’t go any faster, evenfor the most practiced practitioners.
> The emphasis of the important PCT insight that within the individual”it’s all perception” in contrast to the naive objectivism of, say, cognitivepsychology, should not be exaggated into a solipsist neglect of the externalenvironment.
I think I’m safe from that accusation. I probably spend 90 percent of thetime required to produce a model like the Little Man in constructing arealistic physical model of the environment with which the system interacts.The actual control-systemmodel is trivially simple in comparison. Even in our simple tracking models, weinclude more in the environment than most other modelers do –for example, disturbances that directly affect the outcome, and oftennonlinearities and changes in parameters.
> The question of how and under what conditions people can achievecertain “objective” performance, in other words, what guarantees that a systemorganized in a certain way survives (“objectively”) in a given environment,deserves more attention and better answers than “otherwise, reorganization willcontinue”.
It certainly does, and we have never used reorganization to make up forlack of specificity in a model. We match models to performance by adjustingparameters, thus answering the question as to what guarantees success (equal tothat of the human) in the task. “Survival” isn’t so much of a concern; so farall of our experimental subjects have survived. Most of our talk aboutreorganization is by way of speculating about how we would go about includingit in a model, when we finally get around to doing that.
Of course when we’re just doing shirtsleeve conjecturing or Big Picturefantasizing, anything goes. It’s all reorganization. Heck, maybe it’s allchocolate syrup.
> … but still some neurophysiological evidence about basic propertiesof nerves and muscles can replace this data and prove the uselessness offeedback for certain kinds of actions.
I suspect that it would also prove the uselessness of supposing that suchactions can actually occur. Given such data, which I certainly used a good dealof in designing the Little Man, we can find out how good a control system canbe built around those properties, and that is in fact what I did. I learnedenough to know that I am not going to waste my time trying to devise a centralpattern generator that could reproduce the same performance within even afactor of ten worse. Anyone else is welcome to try. I wish someone would. Thenwe could drop this whole stupid subject of open-loopbehavior.
Avery Andrews (930130.0420) —
> My perception is that the next issues to look at are (a) pattern-generators(b) the idea that the output of the effectors is what is controlled (as in theJack Adams quote, and Abbs & Winstein’s `technical’ definition of feedback.This is sort of like what you call the `objectification blunder’ in QAPR, but Iseem to want to call it the `output blunder’.
Yes. I really appreciate all the time, research, and thought you’ve givento PCT in your off-season,so while I agree that these are important subjects I don’t mean to imply thatyou’re expected to go on doing all the work. Linguistics calls, I know.
Pattern generators aren’t a big problem, actually. It’s true that a patterngenerator can always be devised to duplicate the performance of a controlsystem in a specific situation, at least over a brief period of time. But whenyou consider the data that must be available to the generator, and the accuracywith which calculations must be carried out, and the fact that physiologicalmachinery like nerves and muscles change with use, and that there areindependent and unpredictable disturbances that can act directly on the outcomeand that can’t be sensed or anticipated, and that behaviors can last for hoursand days, each move beginning where the last one left off –the whole idea begins to look highly impractical. The simplest behaviorrequires a supercomputer to carry it out.
The pattern generator can be part of a control system, as you’ve noted, butthat’s not the usual idea. The normal open-looppattern generator is actually ruled out because the output blunder must beruled out. Pattern generators rely on output devices that behave uniformly andwith infinite precision, and on total absence of disturbances that enter theoutput chain after the effectors. The only way that I know of to deal witheffectors of variable properties and independent disturbances downstream fromthe effector is through feedback control. There’s just nothing else that willwork. So the output blunder is really the most important one.
This blunder isn’t quite so easy to find in the literature, not because itisn’t there but because it’s like a black hole in the middle of an explanation:invisible unless you realize that something ought to be there. Consider theidea of an “orienting response.” This is a response that orients the body orhead toward a stimulus like a sound. But what kind of response could make theangle of the head end up in exactly the right direction regardless of theorientation of the body or the location of the stimulus? On one occasion thehead might have to turn 2 degrees on the neck; on another, 90 degrees; on stillanother, 40 degrees the other way. How can a response be so specifically whatis objectively required, despite large differences in initial conditions? Thewhole problem is wrapped up and concealed in that little word, “toward.”
There are lots of cases in which responses are simply named after theeffect that the organism’s actions produce, thus skipping the entire questionof how the nervous system can cause the same remote consequence to occur overand over under varying conditions. Just by calling all outcomes of actions”responses” you can leap right over the black hole without giving it a glance.The shoelace-tyingresponse. The problem-solvingresponse. The balancing response. The verbal response. The tracking-the-targetresponse. The putting-it-on-the-Visa-cardresponse. The car-steeringresponse. You can take it for granted that any time the word “response” isused, it’s really referring to an outcome, not the action that happens to berequired this time to create that outcome.
The output blunder is hard to detect because hardly anyone even realizesthat there’s a problem here. It’s not discussed one way or the other. What’swrong with saying that you respond to a question by stating the answer?
Date: Fri Jan 29, 1993 8:58 pm PST
Subject: Powers on (not so) Slow feedback
[from Gary Cziko 930130.0430 GMT] Bill Powers (930128.2200)elucidated:
> Imagine a control system that has a time-constantof 40 milliseconds (we don’t actually have to imagine it; the Little Man modelsit). This control system can control the position of an arm, so that one ormore perceived joint angles follows a smoothly varying reference signal, movingthe arm under complete control at all times.
> But now take this very same control system, and instead of giving it areference signal that passes smoothly from one magnitude to another, make thereference signal jump instantaneously –in, say ten milliseconds –from a frequency of 200 impulses per second to a frequency of 500 impulses persecond, with no passage through intermediate frequencies. . .
> During this maximum-speedmovement, there is naturally no control. The control system is alreadyproducing the largest output it can consistent with stopping in the newposition. Any disturbance that came and went during the approach would simplycause a deviation from the nominal path. So, paradoxically it may seem, thecontrol system’s control action is uncontrolled after a step-changein the reference signal –although the final position is as controlled as ever.
Wouldn’t research to show this be quite easy to do? Simply applydisturbances (big rubber band like I used at Durango) during an action that wasperformed at various speeds. If the HPCT model is right, then for sloweractions the path of the limb will be controlled as well as its final position.At higher speeds, the path will be less well-controlledbut the final position will still be. While at the highest speeds, only thefinal position will be controlled.
Might it take only a big rubber band and a well-placevideo camera to do this research? (A more sophisticated way of introducingdisturbances would be a knee or elbow brace with the physical resistance of thejoint manipulable by remote control–inyour spare time, Bill).
But for some reason I have the intuition that at the very highest speedseven the final position will not be well-controlled,at least not without some patch-upcorrection at the end of the movement. But I want my intuitions to be wronghere.
Looks like a good project for a master’s thesis.–Gary
P.S. It’s possible that my College of Education might in the near futureinherit our campus’s Department of Kinesiology. Could be fun.
Date: Sat Jan 30, 1993 12:41 pm PST
Subject: feedback too slow –stack
[From Rick Marken (930130.1200)]
Avery Andrews (930129.1823)
> Well, I’m just not game to go around claiming that Houk & Rymer,or P.M.H. Rack, or Abbs & Winstein don’t know what they’re doing. If youare, go ahead.
I’m certainly willing to go around claiming (if it is true) that they don’tknow doggie breath about control theory. I’m one of those protestant typescientists that Bill just mentioned –I’ve just got no respect for the priesthood (or the rabbinate, for thatmatter).
Bill has gone to some lengths to explain how “reaction time” might fit intothe behavior of a control system. I plan to make a HyperCard stack that willillustrate the “feedback too slow” issues that are involved when you areactually dealing with a control system.The stack will be based on a singlecontrol system; the user should be able to vary the time constant and transportlag of the system; the user should also be able to apply an impulse or stepdisturbance to the controlled variable. The user should also be able to selecta fixed reference input or one that shifts between two values at a selectedtime rate. Any other suggestions for such a stack? The idea is to show whatwould happen in a typical “reaction time” experiment if the subject were acontrol system. Bill has explained this very clearly but maybe it would help toactually see it dynamically (I know this can be done with the Little Man Demobut the stack might help isolate and clarify the problems with the idea that”feedback is too slow” in a control system). If I actually succeed in buildingsuch a stack maybe Avery could use the results of experiments on the stack inhis critique of the “feedback too slow” myth.
Date: Sat Jan 30, 1993 6:26 pm PST
Subject: slow feedback, output error, etc.
[Avery Andrews 930130.1100 (mostly Bill Powers (930129.1300))
Feedback too slow:
My understanding was that the famous left jab *takes* 40ms from start tofinish. If this is true, it clearly can’t be done by tracking a step-changein a position reference-level.I don’t think it’s true that you necessarily either have a control system (allthe time) or not -just run the positional information through an inhibitable interneuron on itsway to the comparator, & you can switch off the feedback by turning on theinhibition. & it doesn’t disturb me that you would need a pretty fancycircuit element to generate the commands for these movements -after all, it take a tremendous amount of practice to acquire such things. Andthere’s all this talk about `refractory periods’ and the like which could beevidence for various kinds of inhibitions being switched on and off.
& I’d agree that `for what’ is the right thing to say if someone says`feedback is too slow’ & you want to respond brusquely, but I thought myslightly subtler way of saying the same thing would be more useful. After all,we don’t want PCT to be taken up by people who just believe what we say, but bypeople with some capacity to draw their own conclusions from whatever happensto be in front of them. You don’t have to say it all at once, and often it’sbetter not to, I think.
My line on them is that they’re just irrelevant to the question of what therole of feedback is. People thought they were a problem because they confusedfeedback with peripherally mediated response chaining. What CPGs normally dueis specify perceptual reference level contours. From what little I’ve readabout gait control in insects, it seems to involve a complex mix (differentfrom species to species) of CPGs, response-chainingeffects, and actual control. I expect Adams (1971) to be a very informativesource on this subject, but we shall see.
We’re using the term in two different ways. Your `output blunder’ is thebelief that there are effectors that just produce the results intended. Mineis a mistaken idea about what feedback means, the assumption that feedbackmeans monitoring something `directly’ produced by the effectors (an incoherentnotion, I would say, but people really do seem to believe in it). I think thisis a critical mistake, which, for example, gives us the mumbo-jumboof `coordinative structures’ rather than feedback control of relationships.There may be a better term for the blunder, but it’s got to be a snappy one,since it’s so fundamental. Maybe I’m just getting carried away by enthusiasm,but I’ll go so far as to suggest that it may be the *most important* of theblunders.
Kugler et. al.:
One thing that would be extremely useful is for somebody who really knewtheir differential equations to look into Turvey, Kugler et. al. I can’treally be sure whether they are saying profound and useful things (while beingthoroughly dishonest in their portrayal of what other people are doing), orwhether they are just making straightforward things look deep, dark anddifficult by putting them in the most abstract mathematical setting they canfind (e.g. is `limit oscillator’ just another way of talking about anoscillator whose output is a perceptual reference level). My nose tells me thatthere’s more than a little of the latter in their story, but my math is tooweak to be sure.
One of the reasons the linguistics discussions are so convoluted is thatpeople want to talk about the subject before having learned much of anythingabout it, and it isn’t plausible for Bruce or I to try to run a basiclinguistics course on CSGNet. I don’t have any problem with the stuff that RayAllis said, or that various other people said later, but it doesn’t have muchto do with what linguists are talking about when they talk about something like`the structure of the lexicon’.
Linguistics is mostly about patterns.
Unfortunately, most of the verbal patter that goes along with linguisticsis nonsense, so that critically minded people without an aptitude forperceiving the patterns get bogged down in the nonsense, and never get anysense of what is going on (generative grammarians are perfectly happy with theidea of distributed representation of roots in the lexicon, for example,especially because the properties of `strong verb tense stem formation (dig vs.dug, sing vs. sang vs. sung) are highly consistent with what you’d expect fromdistributed representations). If I thought that sorting out linguistics was ahigh priority for PCT, I’d spend more time trying to make it look sensible, butI just don’t think it’s the best use of my time. Something a lot morepromising in the intermediate term would be PCT phonetics and then phonology,since some of the work on `distributed compensatory responses’ is actuallyabout articulatory phonetics. & I’ve never seen an introductory phoneticsbook with a coherent discussion of the difference between happening to producea sound and successfully controlling for producing it under a variety ofcircumstances.
Date: Sat Jan 30, 1993 6:26 pm PST
Subject: slow feedback
[Avery Andrews 920130.1245] Rick Marken (930130.1200)
> I’m certainly willing to go around claiming (if it is true) that theydon’t know doggie breath about control theory. I’m one of those protestant typescientists that Bill just mentioned –I’ve just got no respect for the priesthood (or the rabbinate, for thatmatter).
Neither do I, but in the preliminary stage where you’re basically justtrying to get people to take you seriously, it’s important to start with themost plausible sounding claims, & I doubt that claiming that Rack, Winsteinal. are ignorant about feedback is one of these. It seems to me that there aremany *much* softer and juicier targets around. For example, I found a passageby Feldman and Berkenblit (JMB:20:369-373)where they clearly show that they think that the only alternative to suddenreset of an equilibrium point is to compute a series of intermediate EPs andthen present them to the motor system at a high rate.
I’d be interested in the Hypercard stack when its done -I’ve actually got a Mac on my desk, but I’ve been ignoring it since its notnetworked. This is supposed to change soon, & when it does I’ll pay moreattention to it (and probably try my hand at Think C programming). I’ve got ahalf-finishedjoint simulator, but its treatment of muscles isn’t any good yet, so I can’treally demonstrate anything with it.
Date: Sun Jan 31, 1993 9:02 am PST
Subject: FB 2 Slow; Misc neurology
[From Bill Powers (930131.0900)] Avery Andrews (930130.1100) —
> My understanding was that the famous left jab *takes* 40ms from startto finish. If this is true, it clearly can’t be done by tracking a step-changein a position reference-level.
Let’s plot radius of fist from shoulder as a function of time:
| ref radius
| | | *
| | | *
| | | *
| | —-*|——————————————
| | * | ^radius of opponent’s belly^
| | * |
| | * | <—t.c. = 40 msec
| | * |
| | *
| | * starting radius
The time constant, in a linear system, is independent of the
amplitude of the movement. In the real system, which is somewhat
nonlinear, this is not quite true.
I have drawn the above diagram so that the reference position is set beyondthe point of contact just enough so the fist reaches the opponent’s gut one time-constantafter the start of the movement, or 40 milliseconds.
Note that if Ali set the reference position AT the point of contact, thefist would touch the surface gently at zero velocity after 120 to 160milliseconds. Prize fighters are taught to aim to hit a point well beyond theactual target. They develop large muscles not to push harder, but to acceleratethe fist faster under the impetus of the large initial error signal that iscaused by the step-changein reference signal.
> I don’t think it’s true that you necessarily either have a controlsystem (all the time) or not -just run the positional information through an inhibitable interneuron on itsway to the comparator, & you can switch off the feedback by turning on theinhibition. & it doesn’t disturb me that you would need a pretty fancycircuit element to generate the commands for these movements -after all, it take a tremendous amount of practice to acquire suchthings.
It’s not fair to posit undiscovered connections behaving in unobserved waysto explain a phenomenon that’s adequately explained by a much simpler systemusing known connections behaving in known ways. And if it’s speed you’re after,why add more stages of neural processing?
As I’ve tried to point out, switching off the feedback would remove a greatdeal of the damping in the arm system, so instead of getting a graph like theabove one you would get an underdamped oscillation of position that might takefour or five dimininishing oscillations to come approximately to rest.
The muscle in the above situation is given an initial driving signalexactly equal to the amount of step-changein the reference signal (because the negative feedback takes time to build up).In the first 10 or 20 milliseconds, there is practically no difference betweenthe behavior of the control system and the behavior we would see if thefeedback paths were cut. The difference is the one between the curve and astraight line having the same initial slope. The initial slope represents theacceleration of the arm’s mass under a sudden maximal muscle force.
When you get rid of the feedback, you might gain a very slight amount ofspeed, the difference between the exponential curve and the straight-lineextension of the initial slope. What you would lose would be control of the end-point–for example, the ability to stop the arm quickly and bring the fist back aftera miss.
Hang a small weight like a cup from two or three rubber bands strungtogether in series and see how fast you can control the height of the weightabove a table with no feedback –just by moving the end of the rubber band to a precalculated position. That’show an arm behaves with no feedback.
> And there’s all this talk about `refractory periods’ and the likewhich could be evidence for various kinds of inhibitions being switched on andoff.
The “refractory period” is the time it takes for a neuron to recover afterfiring before the next impulse can be generated. It is on the order of onemillisecond, and is shorter if the neuron is driven by a high-frequencyinput signal. The absolute refractory period –the interval below which no amount of signal will fire the neuron –sets the absolute maximum frequency of firing of the neuron. It is not evidenceof inhibitions being switched on and off.
> People thought they were a problem because they confused feedback withperipherally mediated response chaining.
You keep saying “response chaining.” Doesn’t this convey an image of onecomplete response occurring, setting off another stimulus and another response?This is the image we’re trying to erase, because it implies that while theresponse is in process, the stimulus can’t also be changing. In the realsystem, stimuli and responses are continuous, not alternating.
> From what little I’ve read about gait control in insects, it seems toinvolve a complex mix (different from species to species) of CPGs, response-chainingeffects, and actual control.
Correction: gait control does not involve such things. It is SAID toinvolve such things, on the basis of the only model that people have been ableto think of. People who design models of neural networks have a tendency tomake the neurons into on-offunits, like digital circuits. They know only a few of the actual connections,and those that they do know are treated as binary elements. In the cockroachthere are all kinds of position sensors and rate-of-changesensors, which are simply ignored by modelers like Randall Beer, because theydon’t understand control systems or, for that matter, continuous analogsystems.
> We’re using the term in two different ways. Your `output blunder’ isthe belief that there are effectors that just produce the results intended.Mine is a mistaken idea about what feedback means, the assumption that feedbackmeans monitoring something `directly’ produced by the effectors (an incoherentnotion, I would say, but people really do seem to believe in it).
Right you are. There are really two blunder here: one is the assumptionthat regular consequences can be produced by generating regular motor outputs.The other is the assumption that the controlled variable is the output quantityinstead of the input quantity. Maybe we could distinguish them by calling onethe output-regularityblunder and the other the output-control blunder.
> One thing that would be extremely useful is for somebody who reallyknew their differential equations to look into Turvey, Kugler et. al. I can’treally be sure whether they are saying profound and useful things (while beingthoroughly dishonest in their portrayal of what other people are doing), orwhether they are just making straightforward things look deep, dark anddifficult by putting them in the most abstract mathematical setting they canfind …
I already posted a few comments on Fowler and Turvey’s mathematicalexpertise. I think what happens in most of these cases is that the psychologistteams up with a mathematician; the mathematician just produces the equationsthat the psychologist seems to want, not questioning the rationale, while thepsychologist accepts whatever the mathematician turns out without understandinghow the result depends on the premises. The result looks very impressive,doesn’t it? However, from gobbledygook terms like `limit oscillator’ you cantell who is trying to snow whom.
Linguistics: I think I’ll take a vacation on all that for now.
Best to all, Bill P.
Date: Sun Jan 31, 1993 2:34 pm PST
Subject: Re: Misc neurology
[Avery Andrews 930201.0932] Bill Powers (930131.0900)
Right, I just forgot about the decelerating effect of the opponents body.If Gary’s kinesiologists can be convinced, we might start to get somewhere.Where do they stand on the `motor-action’controversy?
> There are really two blunder here: one is the assumption that regularconsequences can be produced by generating regular motor outputs.
But is this really a blunder? After all, if an `output’ is used as areference level for a control system, you do get a regular result. It is alsomy impression that people have in fact pretty much abandoned the `outputregularity’ blunder for most kinds of movements, at least. This process was theevisceration of the motor program theory, documented in the variouspublications of Schmidt. Once upon a time the idea was that the motor programsspecified alpha-gammaefferents, e.g. real low level motor efference, then they realized that thatwouldn’t work, so there are both feedback and `parameters’, operating in someunspecified way (hence no content to the theory). I suspect it isn’t worthharping on the output regularity blunder any more, tho it certainly would havebeen in the seventies.
Date: Sun Jan 31, 1993 2:44 pm PST
Subject: Re: Ballistics vs. Feedback Demo
[Avery Andrews 930201] (Gary Cziko 930131.1705 GMT)
In the jab story I’ve been pushing, the fist would be propelled the wholeway, so I don’t think your demo makes the point. The throwing with a rubberband attached to your hand one, however, might be useable to very good effect,since it argues, I think, that the actual throwing skill as specified asperceptual reference levels, since otherwise there’s no explanation for thefact that when the elastic is attached, accurate throwing continues to bepossible (a classic motor program would fail).
If this kinesiology department you might absorb is full of frustrated motorprogrammers wondering what to do about Kugler, Turvey et. al., there might be areal opportunity there.
Date: Sun Jan 31, 1993 4:01 pm PST
Subject: Left jab; strategy
[Avery Andrews 930201.1059] Bill Powers (930131.0900)
So, the jab might be driven thru a kinesthetic control system, but it alsomight be that people temporarily deafferent themselves at low levels (rememberthe Bizzi monkey arms), and the fact that it requires so much practice toacquire these moves makes anticipation, etc. much more plausible as a mechanismthan it normally is (as Tom Bourbon just pointed out).
I think it’s important not to spend too much energy on peripheral matterswhere PCT expectations might not be borne out –if you insist that the jab is done via kinesthesis, and somebody proves that itisn’t, then you’ve lost a lot of credit. If you say that it might after all bedone by kinesthesis, then somebody maybe has a nice little hard-edgedresearch project, and while it would be nice if the answer was yes, it wouldnot be a big deal if it wasn’t. It would be quite sufficient to get people tothe point where they could investigate these issues competently -there’s no need to tell them what the answers are going to be.
Date: Sun Jan 31, 1993 7:06 pm PST
Subject: fowler & turvey 1978
[Avery Andrews 930201.1400]
I’ve now got a copy of Fowler & Turvey, & while it’s obvious thatthey missed a lot in their treatment of PCT, there’s also the question ofwhether anyone has written a reorganization demo that solves the problem theypose, or anything like it (the E. coli demo doesn’t count, since you don’t seean actual skill being acquired).
Date: Sun Jan 31, 1993 7:54 pm PST
Subject: fowler & turvey 1978
[From Rick Marken (930131.2000)] Avery Andrews (930201.1400)
> I’ve now got a copy of Fowler & Turvey, …there’s also thequestion of whether anyone has written a reorganization demo that solves theproblem they pose
I did. It’s really easy. I mentioned the results of the simulation in my”Nature of Behavior” paper (first paper in Mind Readings). It’s REALLY easy towrite a PCT simulation of this control system; it is amazinglyefficient.
Date: Sun Jan 31, 1993 10:41 pm PST
Subject: inverse dynamics, fowler&turvey
[Avery.Andrews 930102.1701] Gary Cziko 930201.0402 GMT
I’m not sure what people really think they mean by `ballistic’, but fromwhat I’ve seen, preprogrammed acceleration and deceleration bursts seem to bethe commonest `model’ (for which there is some evidence, in some studiesinvolving yanking on a lever, which is sometimes fixed so that it doesn’t move,but the bursts happen anyway (refereed to by Schmidt, & I just realized Iforgot to write it down & had to take the book back to the library I got itfrom today)). Scare quotes around `model’ since all the stuff that Bill andRick say about the absence of actual models in this area is definitelytrue.
As far as I know, you got kinematics, dynamics & inverse dynamicsright. Inverse dynamics is figuring out what forces the actuators must produceto produce a given motion.
Its interesting seeing how far you can get into a subject like motorcontrol by means of basic scholarship and a bit of computer hacking -surprisingly far, it seems to date. My professional excuse is that peoplesometimes speculate that `motor programs’ were the preadaptation to syntax, sothere’s a reason to investigate them.
(Rick Marken (930131.2000))
Good. On a second perusal of the material, it didn’t look like it would behard to do (and also somewhat deficient as a serious model of skillacquisition). These people do seem to have a bit of a deficit in understandingwritten English.
Date: Mon Feb 01, 1993 12:15 pm PST
Subject: feedback too local
[Avery Andrews 930202.0630]
In the motor control literature I’ve run across statements to the effectthat feedback is `technically’ restricted to monitoring immediate products ofthe effectors (such as, presumably, rpms on a driveshift). Hence, if you aretrying to close your lips and compensate for one being disturbed by moving theother further, this is `technically’ not feedback.
Do any of the engineers on the net recognize this as an actual doctrinefrom courses they took, or anywhere else? Or is it perhaps just an idea thatpsychologists picked up somehow?
Date: Wed Feb 03, 1993 5:52 pm PST
Subject: devils, angels
Today’s find in the library is:
Whiting, H,T.A (ed) _Human Motor Actions: Bernstein Reassessed_, NorthHolland.
Bernstein is one of the culture heroes of Bizzi et. al. on the one hand,and Kugler, Turvey, et. al., on the other, and appears to have been a majorRight Thinker, having, for example, a very nice feedback diagram (pg. 358, alsopg. 130 of N. Bernstein 1976 _The Coordination and Regulation of Movements_,Pergamon press).
The devil’s bib entry is Kugler and Turvey `An Ecological Approach toPerception and Action’ pg. 373-412,esp. pg. 391-392,where they criticize Bernstein’s and everybody else’s conception of feedbackcontrol on various grounds, including what appears to be the idea that itinvolves `an orderly sequence of symbol strings (the representational formatfor the quantities and the commands)’. They appear to be making the event-basedblunder and several others besides, and to be criticing feedback control ingeneral on the basis of somebody’s proposal for a symbolic `motor program’ tocontrol walking (due in its original form to MacKay, W.A. (1980) `The motorprogram: back to the computer’, Trends in Neuroscience 3:97-100). I’m getting the impression that part of their strategy is to criticize actualproposals on the basis of the most bungled derivatives of those proposals thatthey can find (a tactic we should be careful to avoid).
On the other hand, in the next section, they make what strikes me as asensible argument that processes that may look like feedback control w.r.t. aset-pointaren’t necessarily so, and that respiration rate-stabilizationin fact isn’t.
Then there’s an interesting-lookingargument by G. Hinton arguing that it is in fact a good idea to precomputetorques, and discussing various features of muscles that make this easier to dothan you would think (for example, the viscosity properties of the force-velocityrelations give you what is in effect instant feedback control over the velocityof a limb: if the velocity is less than expected, the force exerted by themuscle will be higher, so the velocity lag will be corrected. If Bill Powersand Greg Williams haven’t thought about this article, they ought to.
Date: Thu Feb 04, 1993 9:40 am PST
Subject: control article
[From Rick Marken (930204.0900)]
Hans Blom (930204) —
> William S. Levine, Gerald E. Loeb’s paper ‘The Neural Control of LimbMovement’
> ‘ the study of biological systems should not be confined to testingwhether their performance is compatible with control schemes invented todate but must include detailed examination of their inner workings todiscover new types of control’.
What do they mean by ‘new types of control’? There is only one type thatI’m aware of –maintenance of a perceptual variable at a specified reference level in thecontext of variable disturbances. Levine and Loeb give no evidence ofunderstanding that it is a perceptual variable that is controlled (by whatevermeans –bang/ bang, continuous output, lagged output, etc) and they seem to assume thatthey already know which biological variables are con-trolled; the only problem (they think) is to figure out the mechanism by whichcontrol is implemented. Levine and Loeb (didn’t they have a run in with a youngkid in Chicago some years back?) seem to have forgotten to mention step one inthe study of biological control systems –the test for the controlled variable. How can they compare the performance of aliving system to know “control schemes” if they don’t know what is the livingsystem is controlling? Do they explain how they know what variables arecontrolled by the high jumper that they mention? If so, how do theyknow?
Sounds like another nice entry for the Devil’s Bibliography.
Date: Thu Feb 04, 1993 3:34 pm PST
Subject: telephoning devils
[Avery Andrews 930205.0938] (Greg Williams 930204)
Well, my internal model of Real Professors tells me that publications aremuch more likely to have an effect than phone calls (e.g. some chance, ratherthan no chance at all). Partly because the target has some time to think aboutwhat has been said, and come up with a genuinely useful response (I think thisis much more time-consumingthan the conventional rules for spoken debate allow for), partly because theyhave a strong motive to respond (so as not to be humiliated in public), andpartly because you then get a chance to make an impression on the milling crowdof uncommitteds, who are the guys you’re supposed to win over in order to pulloff a scientific revolution.
Date: Wed Feb 17, 1993 7:10 pm PST
Subject: I’m a thermostat
[FROM: Dennis Delprato (930217)]
No restraints or sanity hearings needed, for….
“What’s it like to be a thermostat (as opposed to being talked about beinga thermostat)?”
Once, again I am reminded of the problem created when control systems aretalked about from an outside perspective:
>Bill Powers (930217.1030)
> As a … third-partyobserver you can say [‘The signal in my home thermostat is about thetemperature in my house’]. As a thermostat, you could not say it. ….
How is it that despite the many depictions of heating and cooling systems,complete with thermostat, in today’s literature, virtually everyone describeswhat’s going on as control of output? I suggest someone prepare a kindlylittle essay (?) spelling out how easy it is to be deceived when one looks fromthe outside in and even does a bang-upjob of describing what they observe. Point out how one gets a very differentpicture when one “takes the viewpoint of the thermostat.” Seems like PCT isgetting closer and closer to Stephenson’s Q-methodologicalthinking that stresses how psychology ignores the person’s point of view,instead imposing the observer’s point of view on the person and calling itunderstanding the person.
Bill reminded me that even the thermostat “has a point of view,” and thisis what he (Bill) is concerned with.
Date: Tue Mar 09, 1993 5:46 am PST
Subject: ANN. REV. PSYCHOL. atricle
From Greg Williams (930309)
New reference to PCT and related work: Paul Karoly, “Mechanisms of Self-Regulation:A Systems View,” ANNUAL REVIEW OF PSYCHOLOGY 44, 1993, 23-52.Even cites LIVING CONTROL SYSTEMS! Wow!! And claims that it was edited by R.S.Marken!!! Ouch!!!! Oh, well, I suppose Rick could use some academic browniepoints. I surely don’t need any. The bottom line on Karoly: lots of devil’sbibliography quotes. For example (p. 30):
Based upon a century-oldinsight attributed to William James (cf Powers 1989), that humans are ‘unique’in nature because they can produce consistent ends by variable means, a numberof contemporary (post-1960) models of dynamic self-regulationhave been developed under the imprimatur of cognitive theory, control/systemsscience, cognitive social learning, or European action theory. All presume thaton-lineregulation is a dynamic process, continuous and holistic rather than linear,built upon the operation of feedback [now hang on] (knowledge of results) andfeedforward (stand-produceddisequilibrium) [say what?] , sensitivity to action-producedenvironmental changes, the accessibility of goal representations [?], and acapacity for the selective mobilization of energy [damn straight, if they’regoing to move around!], attention, and relational judgment.
And so it goes….
As ever (even as Pat and the kids STILL have the flu –remarkable!),
Date: Thu Apr 15, 1993 6:33 am PST
Subject: THE SAME DAMN LIES
[FROM: Dennis Delprato (930415)]
I made the mistake of picking up the most recent issue of the Annual Reviewof Psychology (where we find the state-of-the-artof self-regulationput forth). Once again I am reminded (and not only by ONE chapter) of anepisode J. R. Kantor related.
After he retired from Indiana Univ., he once spent some time at JohnsHopkins where he met with graduate classes. On one occasion, to help studentsdistinguish between what he was saying and conventional views, he asked them tobring their text to class (turned out to be one of the classics in sensation–Geldard’s). At the next meeting, he proceeded to point out unjustified fundamentalassumptions and alternative ways of describing what is happening. The resultwas that the students said this was the first they heard of an alternative toGeldard’s accounts. They rejected Kantor’s ideas out of hand and the overalloutcome of the episode was “a pretty bad scene.” Kantor goes on, “They wereall close to their Ph.D. And what are they going to do, GO OUT AND TELL THESAME DAMN LIES.”
I imagine that anyone who truly appreciates what PCT is all about at themost general level, will time and time again have occasion to think back toKantor’s run-inwith some of today’s leaders and the teachers of those who are beginning totake over the strange discipline of psychology.
I can’t help think of the now best-forgotten”arrogance” thread of the past month or so on CSG-L. But when I read statements of the sort mainstream psychologists are wont tomake, “arrogance” comes to mind. Other descriptors are “ignorance,””illusory,” and “delusory.” What passes for the social sciences, in general,is an interesting group of disciplines. The major criterion for widespreadacceptance of even the most far-fetchedideas is that they be presented in authoritative ways. So-calledscience is simply used as a conservative force to perpetuate tradition. I tryto be “accepting” of divergent views, but will draw the line when the views arenot presented as tentative and open to test, and instead come out as downrightLIES.
Dennis Delprato Dept. of Psychology Eastern Michigan Univ.
Ypsilanti, MI 48197 email@example.com
Date: Tue May 04, 1993 10:42 am PST
Subject: Draft for joint paper
[FROM Dennis Delprato (930503)]
I’d like to see an attempt made for BBS. I might be able to serve a usefulfunction in the form of a critic who looks at the ms. from the standpoint of an”outsider.” PCT has so much to offer–inpresenting it to others, one has to be sensitive to their point of view.”Selling” new ideas isn’t easy, but is a challenge. There certainly need be no”compromises.” One must, however, take into account where readers are comingfrom. This is one reason why one major section should consist of directlyaddressing objections that one finds in the literature. Also, some account ofthe evolution of PCT early on is important. As regards the latter, “systems” isno longer an out construction. Indeed systems are in, and a presentation shouldaddress the relationship between PCT and systems in a positive way.
Tom, could you pass on the previous objections to me?
I imagine you all are aware of the many hoops Harnad poses. I recentlyserved as one of the reviewers of an initial submission that was presented lastJuly, I believe. The author had to address a long list of objections from ninereviewers and the ms. is only now in its second stage, having been sent back tothe reviewers. This is common for papers that eventually get published.
Date: Tue May 25, 1993 5:31 am PST
Subject: Another Devils’ Bib. entry
From Greg Williams (930525)
Quoted from Edwin A. Locke (University of Maryland) and Gary P. Latham(University of Washington), A THEORY OF GOAL SETTING & TASK PERFORMANCE,Prentice-Hall,Englewood Cliffs, New Jersey, 1990, pp. 19-23.(Copyright 1990 by Prentice-Hall,Inc.)
According to the book’s index, there are no other comments on PCT besidesthese.
“As the influence of behaviorism has declined, a neo-behavioristtheory is emerging to take its place. It is called control theory and can beviewed as a combination or integration of behaviorism, machine-computertheory (cybernetics), goal setting theory [championed by Locke and Latham],and, by implication, drive-reductiontheory. It is derived directly from Miller, Galanter, and Pribram’s TOTE model(1960). The major concepts of control theory have been presented by Campion andLord (1982), Carver and Scheier (1982), Hyland (1988), Lord and Hanges (1987),Powers (1973), and others. In brief, the theory asserts that there is INPUT (astimulus), which is detected by a SENSOR. If there is a deviation (also calleda ‘disturbance’), a SIGNAL is sent to an EFFECTOR, which generates modifiedOUTPUT (a response). This output becomes input for the next cycle. In goaltheory language, the input is feedback from previous performance, the referencesignal is the goal, the comparator is the individual’s conscious judgment, andthe effector or response is his or her subsequent action which works to reducethe discrepancy between goal and performance.
“While control theory acknowledges the importance of goal setting, thereare serious, if not irredeemable, flaws in the model. First, observe that themajor ‘motive’ for action under control theory is to remove disturbances ordiscrepancies between the goal and the input (feedback). The natural state ofthe organism is seen to be one of motionlessness or rest. This is true ofmachines, but not of living organisms which are naturally active. It is, infact, a mechanistic version of the long-discrediteddrive-reductiontheory (Cofer & Appley, 1967). Nuttin (1984 [J. Nuttin, MOTIVATION,PLANNING AND ACTION, Erlbaum, Hillsdale, New Jersey]) has observed that in thisaspect, control theory fundamentally misstates the actual source of motivation:’The behavioral process… does not begin with a “test” of the discrepancybetween the standard and the actual states of affairs. Instead, it begins witha preliminary and fundamental operation, namely the construction of thestandard itself, which, as a goal, is at the origin of the action and directsits further course’ (p. 145). Similarly, Bandura (in press [A. Bandura,”Reflections on Nonability Determinants of Competence,” in J. Kolligan & R.Sternberg, eds., COMPETENCE CONSIDERED: PERCEPTIONS OF COMPETENCE ANDINCOMPETENCE ACROSS THE LIFESPAN, Yale University Press, New Haven]) noted thatGOAL SETTING IS FIRST AND FOREMOST A DISCREPANCY CREATING PROCESS. Controltheory begins in the middle rather than at the beginning of the motivationalsequence. To quote Bandura (in press):
Human self-motivationrelies on both DISCREPANCY PRODUCTION and DISCREPANCY REDUCTION. It requiresFEEDFORWARD control as well as FEEDBACK control. People initially motivatethemselves through feedforward control by setting themselves valued challengingstandards that create a state of disequilibrium and then mobilizing theireffort on the basis of anticipatory estimation of what it would take to reachthem. After people attain the standard they have been pursuing, they generallyset a higher standard for themselves. The adoption of further challengescreates new motivating discrepancies to be mastered. Similarly, surpassing astandard is more likely to raise aspiration than to lower subsequentperformance to conform to the surpassed standard. Self motivation thus involvesa dual cyclic process of disequilibrating discrepancy production followed byequilibrating discrepancy reduction. (p. 23 of preprint)
“Figure 1-3[not reproduced here] shows how little of the motivational process controltheory, in its ‘core’ version, incorporates.
“The above is important because if discrepancy reduction is the majormotive, as implied by control theory, then the most logical thing for anindividual to do would simply be to adapt his or her goal to the input. Thiswould guarantee that there would be no disturbance or discrepancy. Machines, ofcourse, cannot do this because the standard has been fixed by people at acertain level (as in setting a thermostat). But people can and do changestandards that diverge from present performance. If the individual’s majormotive were to remove disturbances, people would never do this. Controltheorists argue that lower-levelgoals are actually caused by goals at a higher level in the individual’s goalhierarchy (Carver & Scheier, 1982). But this only pushes the problem back astep. Why should people set higher-level goals if they only want to reduce tension? But in reality, people do setgoals and then act to attain them; they do not focus primarily on eliminatingdisturbances. Removal of discrepancies and any associated tension is aCORRELATE of goal-directedaction, not its cause. The causal sequence begins with setting the goal, notwith removing deviations from it.
“At a fundamental level, discrepancy reduction theories such as controltheory are inadequate because if people consistently acted in accordance withthem by trying to eliminate all disturbances, they would all commit suicide –because it would be the only way to totally eliminate tension. If people choseinstead to stay alive but set no goals, they would soon die anyway. By the timethey were forced into action by desperate, unremitting hunger pangs, it wouldbe too late to grow and process the food they would need to survive.
“In their major work, Carver and Scheier (1981) denied that discrepancyreduction is motivated by a desire to reduce a drive or state of tension. Buttheir own explanation as to why people at to reduce discrepancies is quitepuzzling. ‘The shift [of action in the direction of the goal or standard] is anatural consequence of the engagement of a discrepancy-reducingfeedback loop’ (p. 145). This statement, of course, explains nothing. Why isdiscrepancy reduction a ‘natural consequence’? According to goal theory, BOTHdiscrepancy creation AND discrepancy reduction occur for the same reason:because people need and desire to attain goals. Such actions are required fortheir survival, happiness, and well-being.
“A second problem with control theory is its very use of a machine as ametaphor. The problem with such a metaphor is that it cannot be taken tooliterally or it becomes highly misleading (e.g., see Saundelands, Glynn, &Larson, 1988 [L.E. Sandelands, M.A. Glynn, & J.R. Larson, “Task Performanceand the ‘Control’ of Feedback,” Columbia University, unpublished manuscript]).For example, people do not operate within the deterministic, closed-loopsystem that control theory suggests. In response to negative feedback,forexample, people can try harder or less hard. They can focus on the cause andperhaps change their strategy. They can also lower the goal to match theirperformance; in some cases they may raise their goal. Furthermore, they canreinterpret the discrepancy as unimportant and ignore it or can even totallydeny it. They can also question the accuracy of the feedback. They can gooutside the system (by leaving the situation). They can attack the person theyhold responsible for the discrepancy. They can become paralyzed by self-doubtand fear and do nothing. They can drink liquor to blot out the pain. In short,they can do any number of things other than respond in machinelike fashion.Furthermore, people can feel varying degrees of satisfaction anddissatisfaction, develop varying degrees of commitment to goals, and assesstheir confidence in being able to reach them (Bandura, 1986). These emotions,decisions, and estimates affect what new goals they will set and how they willrespond to feedback indicative of deviations from the goal (Bandura, 1988).Control theory, insofar as it stresses a mechanistic model, simply has no placefor these alternatives, which basically means that it has no place forconsciousness. Insofar as this is the case, the theory must fail for the samereason behaviorism failed. Without studying and measuring psychologicalprocesses, one cannot explain human action.
“One might ask why control theory could not be expanded so as toaccommodate the ideas and processes noted above. Attempts have been made to dothis, but when it is done, the machine language may still be retained. Hyland(1988), for example, described the effects of goal importance or commitment interms of ‘error sensitivity,’ which is represented diagrammatically by a boxcalled an ‘amplifier.’ Expectations and memory are represented as ‘symboliccontrol loops.’ Decision making is done not by a person but by a ‘selector.’What is the benefit of translating relatively clear and well-acceptedconcepts that apply to human beings into computer language that is virtuallyincomprehensible when used to describe human cognition? The greater the numberof concepts referring to states or actions of consciousness that are relabeledin terms of machine language, the more implausible and incomprehensible thewhole enterprise becomes. Nuttin (1984, p. 148) wrote on this: ‘When behavioralphenomena are translated into cybernetic and computer language, theirmotivational aspect is lost in the process. This occurs because motivation isforeign to all machines.’
“On the other hand, if additional concepts are brought into controltheory and not all relabeled in machine language (e.g., Lord & Hanges,1987), then control theory loses its distinctive character as a machinemetaphor and becomes superfluous –that is, a conglomeration of ideas borrowed from OTHER theories. And if controltheory does not make the needed changes and expansions, it is inadequate toaccount for human action. Control theory, therefore, seems to be caught in atriple bind from which there is no escape. If it stays strictly mechanistic, itdoes not work. If it uses mechanistic language to relabel concepts referring toconsciousness, it is incomprehensible. And if it uses nonmechanistic concepts,it is unoriginal. It has been argued that control theory is useful because itprovides a general model into which numerous other theories can be integrated(Hyland, 1988). However, a general model that is inadequate in itself cannotsuccessfully provide an account of the phenomena of other theories.
“In their book, Carver and Scheier (1981) examined the effect ofindividual differences in degree of internal focus versus external focus inaction. While this presentation is more plausible than the mechanistic versionsof control theory, most of it actually has little to do with control theory asit relates to goal setting. For example, they discuss how expectancies and self-focusaffect performance but do not examine the goal-expectancyliterature (as we do in Chapter 3). And some of their conclusions (such as thatself-efficacydoes not affect performance directly) contradict actual research findings. Onlyone actual goal setting study (not in Carver and Scheier’s book) has used theself-focusmeasure. Hollenbeck and Williams (1987) found that self-focusonly affected performance as part of a triple interaction in which ability wasnot controlled. Thus it remains to be seen how useful the measure is, either asa moderator or as a mediator of goal setting effectiveness.
“There is also a conceptual problem with the prediction that the relationbetween goals and performance will be higher among those high in self-focusthan those low in self-focus.Goal attainment requires, over and above any internal focus, an EXTERNAL focus;most goals refer to something one wants to achieve in the external world. Thusthe individual must monitor external feedback that shows progress in relationto the goal in order to make progress toward it. Individuals might focusinternally as well (a) to remind ourselves of what the goal is –though this can also be done externally, as on a feedback chart; (b) to retaincommitment by reminding themselves of why the goal is important; and (c) toassess self-efficacy.Furthermore, depending on what is focused on, (e.g., self-encouragingthoughts or self-doubt),an internal focus could either raise or lower goal-relevanteffort. In sum, the relation between where one is focused and goal-relevantperformance seems intuitively far more complex than is recognized by thecognitive version of control theory.
“Finally, some have argued that control theory is original because itdeals with the issue of goal change (e.g., Campion & Lord, 1982). However,goal change was actually studied first by level-of-aspirationresearchers in the 1930s and 1940s, so control theory can make no claim oforiginality here. Nor can a mechanistic model hope to deal adequately withissues involving human choice as noted above.
“In sum, the present authors do not see what control theory has added toour understanding of the process of goal setting; all it has done is to restatea very limited aspect of goal theory in another language, just as was done bybehavior mod advocates. Worse, control theory, in its purest form, actuallyobscures understanding by ignoring or inappropriately relabeling crucialpsychological processes that are involved in goal-directedaction (these will be discussed in subsequent chapters).”