A memory free control model

[From Bruce Gregory (2010.04.29.1940 EDT)]

To avoid irritating Rick any further, I will not attribute this claim to PCT, which I clearly do not understand. I take responsibility for the following claim which has nothing to do with PCT. I claim that a control model incorporating reorganization but not incorporating any need for memory can explain any behavior that can be modeled by PCT. (In fact, I claim that the vast majority of control models of behavior do not in fact incorporate memory.) If this is incorrect, perhaps someone can rise to this challenge by citing an experiment that cannot be modeled without incorporating memory. If I cannot avoid the incorporation of memory in the model, I will gladly acknowledge the error of my ways.

Rick of course is free to rule me out of order for raising an issue not appropriate for discussion on CSGnet.

Bruce

[From Rick Marken (2010.04.29.1700)]

Bruce Gregory (2010.04.29.1940 EDT)]

I claim that a control model incorporating reorganization but not incorporating any
need for memory can explain any behavior that can be modeled by PCT...If this is
incorrect, perhaps someone can rise to this challenge by citing an experiment that
cannot be modeled without incorporating memory.

How about any memory experiment?

If I cannot avoid the incorporation of memory in the model, I will gladly acknowledge
the error of my ways.

OK. Give it a try. Model the behavior in a free recall experiment
using a control model incorporating reorganization but no memory.
Present the model with, say, 10 lists of 20 words each. Present the
words in each list one word at a time; when all the words in the list
have been presented have the model "write down" all the words in the
list that it can recall. The model should produce the classic "serial
position curve", with pronounced recency and primacy effects.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Rick Marken (2010.04.29.1730)]

Rick Marken (2010.04.29.1700)

Bruce Gregory (2010.04.29.1940 EDT)]

If I cannot avoid the incorporation of memory in the model, I will gladly
acknowledge�the error of my ways.

I just thought of another experiment that is probably better because
it is more obviously a control task. It's Bill's "cognitive control"
experiment, which is basically a compensatory tracking task where the
subject is asked to keep the name of a president (say, Lincoln)
displayed on the screen while a disturbance is acting on the ordered
list of president names, pushing the displayed name to adjacent
presidents. Subjects use a mouse to compensate for the disturbance and
keep the target name (Lincoln) on display. I can write up a version of
this task for you if you would like. Then you could collect some data
and build a model to explain it. This would be a good exercise for
you, I think.

This is a very tough control task -- it might be easier to use a more
familiar ordered set of words, like maybe colors (roygbiv) -- but it
would be very interesting to see if you could model it without memory.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bruce Gregory (2010.04.29.2202 EDT)]

[From Rick Marken (2010.04.29.1700)]

OK. Give it a try. Model the behavior in a free recall experiment
using a control model incorporating reorganization but no memory.
Present the model with, say, 10 lists of 20 words each. Present the
words in each list one word at a time; when all the words in the list
have been presented have the model “write down” all the words in the
list that it can recall. The model should produce the classic “serial
position curve”, with pronounced recency and primacy effects.

BG: I think I understand what I have been missing. It is the Observer that remembers, not the hierarchy. Is that correct? In the same way, it is the Observer that imagines the outcome of an experiment. Memory and imagination belong to the Observer. Action belongs to the hierarchy.

Bruce

[From Bruce Gregory (2010.04.20.2225 EDT)]

[From Rick Marken (2010.04.29.1700)]

OK. Give it a try. Model the behavior in a free recall experiment
using a control model incorporating reorganization but no memory.
Present the model with, say, 10 lists of 20 words each. Present the
words in each list one word at a time; when all the words in the list
have been presented have the model “write down” all the words in the
list that it can recall. The model should produce the classic “serial
position curve”, with pronounced recency and primacy effects.

BG: Good test. The control involves writing down a list of words. Each word is embodied in a control loop where the reference level is the word to be written. Control is good to the extent the reference level for each control loop is a word on the list. If the word is not on a list, the hierarchy experiences an error and reorganizes.

The memory of the hierarchy resides in the individual reference levels of the control loops. (I think I suggested that a few posts ago.)

Bruce

[From Rick Marken (2010.04.20.2230)]

Bruce Gregory (2010.04.20.2225 EDT)]

Rick Marken (2010.04.29.1700)--

OK. Give it a try. Model the behavior in a free recall experiment
using a control model incorporating reorganization but no memory.

BG: Good test. The control involves writing down a list of words. Each word
is embodied in a control loop where the reference level is the word to be
written. Control is good to the extent the reference level for each control
loop is a word on the list. If the word is not on a list, the hierarchy
experiences an error and reorganizes.
The memory of the hierarchy resides in the individual reference levels of
the control loops. (I think I suggested that a few posts ago.)

A working model, please.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bruce Gregory (2010.04.30.0622 EDT)]

[From Rick Marken (2010.04.20.2230)]

Bruce Gregory (2010.04.20.2225 EDT)]

Rick Marken (2010.04.29.1700)--

OK. Give it a try. Model the behavior in a free recall experiment
using a control model incorporating reorganization but no memory.

BG: Good test. The control involves writing down a list of words. Each word
is embodied in a control loop where the reference level is the word to be
written. Control is good to the extent the reference level for each control
loop is a word on the list. If the word is not on a list, the hierarchy
experiences an error and reorganizes.
The memory of the hierarchy resides in the individual reference levels of
the control loops. (I think I suggested that a few posts ago.)

A working model, please.

BG: Could you send me a PDF of the paper containing the data and the PCT model? Thanks.

Bruce

[From Bruce Gregory (2010.04.30.0925 EDT)]

Bruce Gregory (2010.04.29.1940 EDT)]

I claim that a control model incorporating reorganization but not incorporating any
need for memory can explain any behavior that can be modeled by PCT...If this is
incorrect, perhaps someone can rise to this challenge by citing an experiment that
cannot be modeled without incorporating memory.

BG: Ooops! My bad. There is indeed memory in the PCT model. The hierarchy _is_ the memory. Sorry for any confusion I might have caused.

Bruce

[From Rick Marken (2010.04.30.0800)]

Bruce Gregory (2010.04.30.0622 EDT)

Rick Marken (2010.04.20.2230)--

A working model, please.

BG: Could you send me a PDF of the paper containing the data and the PCT
model? Thanks.

and then:

Bruce Gregory (2010.04.30.0925 EDT)--

Bruce Gregory (2010.04.29.1940 EDT)--

I claim that a control model incorporating reorganization but not incorporating any
need for memory can explain any behavior that can be modeled by PCT...If this is
incorrect, perhaps someone can rise to this challenge by citing an experiment that
cannot be modeled without incorporating memory.

BG: Ooops! My bad. There is indeed memory in the PCT model. The hierarchy _is_ the >memory.

No, wrong again.

Sorry for any confusion I might have caused.

You really don't have to apologize to anyone except, possibly, yourself.

Best

Rick

···

---
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.04.30.10.59]

[From Bruce Gregory (2010.04.29.1940 EDT)]

To avoid irritating Rick any further, I will not attribute this claim to PCT, which I clearly do not understand. I take responsibility for the following claim which has nothing to do with PCT. I claim that a control model incorporating reorganization but not incorporating any need for memory can explain any behavior that can be modeled by PCT. (In fact, I claim that the vast majority of control models of behavior do not in fact incorporate memory.) If this is incorrect, perhaps someone can rise to this challenge by citing an experiment that cannot be modeled without incorporating memory. If I cannot avoid the incorporation of memory in the model, I will gladly acknowledge the error of my ways.

You have to define what you mean by "memory". Do you mean memory for a specific event ("Aunt Jane came to dinner last Tuesday"), memory for a usual event ("Aunt Jane always comes to dinner on Tuesdays"), memory for how to do something by rote ("You connect the widget to the frammigus before you lay on the whatmolic"), memory for how to do something ("I know how to ride a bike, but I can't tell you how to do it"), and so forth. In PCT, this last is covered by reorganization, and I think that is much of what Rick is pointing out. The first kind clearly cannot be an effect of reorganization, but the middle two might be, at logical/program levels. I wouldn't put a bet on that, either way.

Martin

[From Bruce Gregory (2010.04.30.1312 EDT)]

[From Rick Marken (2010.04.30.0800)]

BG: Ooops! My bad. There is indeed memory in the PCT model. The hierarchy is the >memory.

No, wrong again.

BG: If that’s the best you can do I won’t bother to respond.

Sorry for any confusion I might have caused.

You really don’t have to apologize to anyone except, possibly, yourself.

BG: That makes about as much sense as your other posts. If you have nothing to say, why not try silence? Or are you simply controlling for having the last word?

Bruce

[From Bruce Gregory (2010.04.30.1320 EDT)]

[Martin Taylor 2010.04.30.10.59]

[From Bruce Gregory (2010.04.29.1940 EDT)]

To avoid irritating Rick any further, I will not attribute this claim to PCT, which I clearly do not understand. I take responsibility for the following claim which has nothing to do with PCT. I claim that a control model incorporating reorganization but not incorporating any need for memory can explain any behavior that can be modeled by PCT. (In fact, I claim that the vast majority of control models of behavior do not in fact incorporate memory.) If this is incorrect, perhaps someone can rise to this challenge by citing an experiment that cannot be modeled without incorporating memory. If I cannot avoid the incorporation of memory in the model, I will gladly acknowledge the error of my ways.

You have to define what you mean by "memory". Do you mean memory for a specific event ("Aunt Jane came to dinner last Tuesday"), memory for a usual event ("Aunt Jane always comes to dinner on Tuesdays"), memory for how to do something by rote ("You connect the widget to the frammigus before you lay on the whatmolic"), memory for how to do something ("I know how to ride a bike, but I can't tell you how to do it"), and so forth. In PCT, this last is covered by reorganization, and I think that is much of what Rick is pointing out. The first kind clearly cannot be an effect of reorganization, but the middle two might be, at logical/program levels. I wouldn't put a bet on that, either way.

BG: Memory of the first and second variety does not constitute behavior, so PCT has nothing to say about it (beyond hand waving). Memory of how to do something by wrote is embodied in the hierarchy as is memory for how to ride a bicycle. They may not have become embodied in the hierarchy as a result of reorganization, but that is where they are must be located if they are to affect behavior.

Bruce

[From Rick Marken (2010.04.30.1130)]

Bruce Gregory (2010.04.30.1312 EDT)--

[From Rick Marken (2010.04.30.0800)]

BG: Ooops! My bad. There is indeed memory in the PCT model. The hierarchy
_is_ the >memory.

No, wrong again.

BG: If that's the best you can do I won't bother to respond.

Sorry for any confusion I might have caused.

You really don't have to apologize to anyone except, possibly, yourself.

BG: That makes about as much sense as your other posts. If you have nothing
to say, why not try silence? Or are you simply controlling for having the
last word?

I don't see why you think it makes no sense. You seemed to be
apologizing for your persistent claims that there is no memory in the
PCT model. So I was just saying that there was no need to apologize to
those of us who are familiar with the PCT model (from reading B:CP
and, sometimes. actually building the models). I suggested that maybe
you just have apologize to yourself for maintaining this self
deception.

The fact that your conclusion about where memory fits into the model
just means that you have a little more work to do at understanding the
model. Reading the Memory chapter in B:CP would be the best place to
start, it seems to me. But you suggested developing a model as a way
of understanding where memory does (or doesn't) fit into PCT and that
really seems like the best idea to me. Words really don't work well as
a modeling tool. That's why I am not interested in having the last
word. I am much more interested in participating in a modeling
exercise.

I proposed modeling the "Cognitive Control" experiment that I
described to you, where the subject is asked to keep the name of a
particular US President on the display while a disturbance acts to
push the display to "adjacent" Presidents. I would make it simpler by
using a simpler ordered sequence of words, like the ordered names of
the colors in the spectrum (per the conventional way of naming those
colors: roygbiv). I think it would be very hard to create a model of
this behavior that did not include memory, not only for the color
names but also for their relationship to each other in the series. If
we worked on building this model I think you would see how memory fits
into a control model. And you would also learn a lot about what we
mean by the term "model" in PCT.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.04.30.15.31]
[From Bruce Gregory (2010.04.30.1320 EDT)]

   

[Martin Taylor 2010.04.30.10.59]

[From Bruce Gregory (2010.04.29.1940 EDT)]

... If I cannot avoid the incorporation of memory in the model, I will gladly acknowledge the error of my ways.

You have to define what you mean by "memory". Do you mean memory for a specific event ("Aunt Jane came to dinner last Tuesday"), memory for a usual event ("Aunt Jane always comes to dinner on Tuesdays"), memory for how to do something by rote ("You connect the widget to the frammigus before you lay on the whatmolic"), memory for how to do something ("I know how to ride a bike, but I can't tell you how to do it"), and so forth. In PCT, this last is covered by reorganization, and I think that is much of what Rick is pointing out. The first kind clearly cannot be an effect of reorganization, but the middle two might be, at logical/program levels. I wouldn't put a bet on that, either way.
     

BG: Memory of the first and second variety does not constitute behavior, so PCT has nothing to say about it (beyond hand waving). Memory of how to do something by wrote is embodied in the hierarchy as is memory for how to ride a bicycle. They may not have become embodied in the hierarchy as a result of reorganization, but that is where they are must be located if they are to affect behavior.

Memory of any kind does not constitute behaviour. A memory like "Aunt Jane came to dinner last Tuesday" is a kind of perception. So is a memory like "Aunt Jane usually comes to dinner on Tuesdays", though it is a perception of a kind different from that of a specific event. And yet another kind of perception is afforded by "You connect the widget to the frammigus before you lay on the whatmolic" -- at least I think it is, though you apparently disagree. Memory that is embodied in the structure of the complex of control units is not a perception, which is why it is nearly impossible to tell someone how to ride a bicycle.

PCT says a lot about perceptions, so it does have something to say about the kinds of memory that are perceptions. Maybe what concerns you is the ill-defined control mechanisms that act to evoke those particular perceptions. It's easy to understand control that works through perceptible environmental affordances (meaning that you can see some effect of the action on the environment and infer its consequent effect on a perception). It's not easy to understand hypothesised actions that generate a desired memory such as "When did Aunt Jane come to dinner". Not knowing how memories are actually stored, or whether memories of different kinds are stored similarly, it is hard to hypothesise testable mechanisms for producing memories that fit the reference values for the corresponding perceptions. What we can say, assuming PCT, is that the evocation of a memory is at least sometimes a controlled process. I do not think it is always a controlled process, any more than are the perceptions generated from sensory data. Like them, some memory perceptions are controlled, most are not.

You guess that rote memory is embodied in the hierarchy. I guess that it isn't. I have no notion of how to test which of these hypotheses is nearer the truth (no more than I have as to whether the pure hierarchy is a viable structure for the complex of control units in a moderately complex brain such as that of a mouse or a fish).

Martin

[From Bruce Gregory (2010.05.01.0707 EDT)]

[Martin Taylor 2010.04.30.15.31]

Memory of any kind does not constitute behaviour.

BG: True.

A memory like "Aunt Jane came to dinner last Tuesday" is a kind of perception. So is a memory like "Aunt Jane usually comes to dinner on Tuesdays", though it is a perception of a kind different from that of a specific event. And yet another kind of perception is afforded by "You connect the widget to the frammigus before you lay on the whatmolic" -- at least I think it is, though you apparently disagree. Memory that is embodied in the structure of the complex of control units is not a perception, which is why it is nearly impossible to tell someone how to ride a bicycle.

BG: I am confused about the nature of perception. You seem to be saying that unless I am conscious of something, it is not a perception. That seems to be at variance with PCT, since I can control perceptions that I am not conscious of controlling. What am I missing?

PCT says a lot about perceptions, so it does have something to say about the kinds of memory that are perceptions. Maybe what concerns you is the ill-defined control mechanisms that act to evoke those particular perceptions.

BG: Yes. I am rereading Chapter 15 of B:CP perhaps that will address my discomfort.

It's easy to understand control that works through perceptible environmental affordances (meaning that you can see some effect of the action on the environment and infer its consequent effect on a perception). It's not easy to understand hypothesised actions that generate a desired memory such as "When did Aunt Jane come to dinner". Not knowing how memories are actually stored, or whether memories of different kinds are stored similarly, it is hard to hypothesise testable mechanisms for producing memories that fit the reference values for the corresponding perceptions. What we can say, assuming PCT, is that the evocation of a memory is at least sometimes a controlled process. I do not think it is always a controlled process, any more than are the perceptions generated from sensory data. Like them, some memory perceptions are controlled, most are not.

BG: My only point was that unless a memory is incorporated into the hierarchy, and its retrieval is therefore controlled, it can not influence behavior. I thought that was part of the PCT model, but I may be mistaken.

You guess that rote memory is embodied in the hierarchy. I guess that it isn't. I have no notion of how to test which of these hypotheses is nearer the truth (no more than I have as to whether the pure hierarchy is a viable structure for the complex of control units in a moderately complex brain such as that of a mouse or a fish).

BG: I was trying not to guess. Rather I was trying to state the PCT position. If behavior is the control of perception and rote memory involves behavior (reciting what was learned) then it seemed to me that rote learning must be incorporated into the hierarchy. Is my logic flawed?

Bruce

[Martin Taylor 2010.05.02.21.25]

[From Bruce Gregory (2010.05.01.0707 EDT)]

[Martin Taylor 2010.04.30.15.31]
     

Memory of any kind does not constitute behaviour.
     

BG: True.

A memory like "Aunt Jane came to dinner last Tuesday" is a kind of perception. So is a memory like "Aunt Jane usually comes to dinner on Tuesdays", though it is a perception of a kind different from that of a specific event. And yet another kind of perception is afforded by "You connect the widget to the frammigus before you lay on the whatmolic" -- at least I think it is, though you apparently disagree. Memory that is embodied in the structure of the complex of control units is not a perception, which is why it is nearly impossible to tell someone how to ride a bicycle.
     

BG: I am confused about the nature of perception. You seem to be saying that unless I am conscious of something, it is not a perception. That seems to be at variance with PCT, since I can control perceptions that I am not conscious of controlling. What am I missing?
   
I'm sorry, but after several rereadings of what I wrote, I am unable to see how you conclude that I assert perceptions to be necessarily conscious. What led you to think I meant that? Each of the examples might or might not be in consciousness at any moment, but it makes them no less a perception whether conscious or not.
   

PCT says a lot about perceptions, so it does have something to say about the kinds of memory that are perceptions. Maybe what concerns you is the ill-defined control mechanisms that act to evoke those particular perceptions.
     

...
   BG: My only point was that unless a memory is incorporated into the hierarchy, and its retrieval is therefore controlled, it can not influence behavior. I thought that was part of the PCT model, but I may be mistaken.
   
I don't connect with your first sentence. As I understand it, if a memory is incorporated into the hierarchy (i.e. in the interconnection structure and the parameterization of the functional elements), it cannot be retrieved. It just IS. Memories that are stored somewhere may or may not be retrieved as the control of perception (finding the particular memery you want for some purpose). They may be retrieved as components of perceptions that are partly based on sensory input, or evoked by the perception of something else (as, for example, the smell of warm raspberry jam once evoked a vision of my aunt's breakfast room in Scotland where I had not been for several decades). There's no control involved in such retrieval processes, so far as I can see.
   

You guess that rote memory is embodied in the hierarchy. I guess that it isn't. I have no notion of how to test which of these hypotheses is nearer the truth (no more than I have as to whether the pure hierarchy is a viable structure for the complex of control units in a moderately complex brain such as that of a mouse or a fish).
     

BG: I was trying not to guess. Rather I was trying to state the PCT position. If behavior is the control of perception and rote memory involves behavior (reciting what was learned) then it seemed to me that rote learning must be incorporated into the hierarchy. Is my logic flawed?
   
I can't see the logic, I'm afraid. Let's try to be a bit more specific, and assume an associative learning task. I say "Edwin" and you say "Targus", and so on. When I say "Edwin", you extract various memories from some storage place, but most of them are not the match to "Edwin", and error persists in teh control process, just as when you want a pencil and you look around, seeing various linear woody things, most of which are twigs or pens or toothpicks, error persists until you actually light upon a pencil. Is the pencil incorporated into the hierarchy? I would guess not, so why would you think that logically a memory developed by rote learning must be?

Try another kind of task. I have learned by rote that before you lay on the whatmolic you must connect the widget to the frammigus. I don't know why you must, but I have learned that it is so. Now I have been told I have to lay on some whatmolic. If I try to do that without connecting the widget to the frammigus, nothing happens. No whatmolic appears. Error! What is it that corrects the error? Is it random reorganization, or is it a search through memories in some way marked as relevant to whatmolic? Is it like finding the pencil or like riding a bicycle without falling down? It seems to me that one of the actions in correcting the whatmolic problem is to provide the appropriate reference value to a memory search control loop.

On the other hand, it does seem as though things learned by rote can, with practice, be incorporated in the hierarchy, so that memory search need no longer be evoked, and controlling for perceiving whatmolic to be laying on properly would directly involve controlling for perceiving the widget to be connected to the frammigus. So I could go along with a modification of your phrase, and say "it seemed to me that rote learning can be incorporated into the hierachy".

Martin

···

On 2010/05/1 7:07 AM, Bruce Gregory wrote:

[From Bruce Gregory (2010.05.05.0800 EDT)]

[Martin Taylor 2010.05.02.21.25]

I can't see the logic, I'm afraid. Let's try to be a bit more specific, and assume an associative learning task. I say "Edwin" and you say "Targus", and so on. When I say "Edwin", you extract various memories from some storage place, but most of them are not the match to "Edwin", and error persists in teh control process, just as when you want a pencil and you look around, seeing various linear woody things, most of which are twigs or pens or toothpicks, error persists until you actually light upon a pencil. Is the pencil incorporated into the hierarchy? I would guess not, so why would you think that logically a memory developed by rote learning must be?

BG: My confusion involved "incorporated into the hierarchy." Now that I have reread Chapter 15 of B:CP is see that I have no problem with the PCT model.

Try another kind of task. I have learned by rote that before you lay on the whatmolic you must connect the widget to the frammigus. I don't know why you must, but I have learned that it is so. Now I have been told I have to lay on some whatmolic. If I try to do that without connecting the widget to the frammigus, nothing happens. No whatmolic appears. Error! What is it that corrects the error? Is it random reorganization, or is it a search through memories in some way marked as relevant to whatmolic? Is it like finding the pencil or like riding a bicycle without falling down? It seems to me that one of the actions in correcting the whatmolic problem is to provide the appropriate reference value to a memory search control loop.

BG: Indeed.

On the other hand, it does seem as though things learned by rote can, with practice, be incorporated in the hierarchy, so that memory search need no longer be evoked, and controlling for perceiving whatmolic to be laying on properly would directly involve controlling for perceiving the widget to be connected to the frammigus. So I could go along with a modification of your phrase, and say "it seemed to me that rote learning can be incorporated into the hierarchy".

BG: Yes, that's along the lines that i was thinking. Thanks.

Bruce