Memory

[From Rick Marken (2010.01.14.1800)]

Bill Powers (2010.01.14.1350 MST)--

BP: I think that is what happens

I think that's what happens, too. But what is happening is memory.
This thread started as a discussion of the closed-loop nature of
behavior in psychological experiments. So I'm changing the subject
line for this conversation to "Memory". If there are any more comments
on the closed-loop nature of behavior in psychological experiments
please use Martin's somewhat misleading but, nevertheless, more
appropriate subject head: Psychophysical studies.

Best

The management

···

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

[From Bruce Gregory (2010.01.15.0255 UT)]

[From Rick Marken (2010.01.14.1800)]

Bill Powers (2010.01.14.1350 MST)--

BP: I think that is what happens

I think that's what happens, too. But what is happening is memory.
This thread started as a discussion of the closed-loop nature of
behavior in psychological experiments. So I'm changing the subject
line for this conversation to "Memory". If there are any more comments
on the closed-loop nature of behavior in psychological experiments
please use Martin's somewhat misleading but, nevertheless, more
appropriate subject head: Psychophysical studies.

Why does any control system need a memory? The thermostat that regulates the heat in my house has no memory but it does a good job controlling the temperature. So long as a control system has an input function and an output function and error detection circuitry, why does it need a memory?

Bruce

[From Rick Marken (2010.01.14.1900)]

[From Bruce Gregory (2010.01.15.0255 UT)]

Why does any control system need a memory? The thermostat that regulates the heat in my
house has no memory but it does a good job controlling the temperature. So long as a control
system has an input function and an output function and error detection circuitry, why does it
need a memory?

Control systems need a memory when they are being used as the basis of
a model of living systems (like people) that have memories.

Best

Rick

···

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

[From Bruce Gregory (2010.01.15.0312 UT)]

[From Rick Marken (2010.01.14.1900)]

[From Bruce Gregory (2010.01.15.0255 UT)]

Why does any control system need a memory? The thermostat that regulates the heat in my
house has no memory but it does a good job controlling the temperature. So long as a control
system has an input function and an output function and error detection circuitry, why does it
need a memory?

RM: Control systems need a memory when they are being used as the basis of
a model of living systems (like people) that have memories.

You’ve built many PCT models. Did any of them incorporate a memory?

Bruce

[From Rick Marken (2010.01.14.1930)]

Bruce Gregory (2010.01.15.0312 UT)--

RM: Control systems need a memory when they are being used as the
basis of a model of living systems (like people) that have memories.

You've built many PCT models. Did any of them incorporate a memory?

Not explicitly, no.

Best

Rick

···

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

[From Bruce Gregory (2010.01.15.1155 UT)]

[From Rick Marken (2010.01.14.1930)]

Bruce Gregory (2010.01.15.0312 UT)–

RM: Control systems need a memory when they are being used as the
basis of a model of living systems (like people) that have memories.

You’ve built many PCT models. Did any of them incorporate a memory?

Not explicitly, no.

Thanks. In that case, I plan to avoid future discussions of memory on CSGnet. If anyone happens to be interested in the neuroscience of memory, I recommend Eric Kandel’s autobiographical In Search of Memory: The Emergence of a New Science of Mind. Kandel won the Nobel Prize in 2000 for his work on the mechanisms of memory. Good textbook treatments can be found in , Principles of Neuro Science by Kandel, et al and From Neuron to Brain, by Nicholls, et al.

Bruce

[From Bruce Gregory (2010.01.15.1220 UT)]

[From Bill Powers (2010.01.14.1350 MST)]

The thing about PCT models is that they tend to go on predicting correctly when you change the experimental conditions. The empirical approach can't do that except by luck.

BG: Could you be a bit more explicit about what you mean by the "empirical approach?" I think of PCT as being empirical. Does "empirical approach" function simply to identify non-control models?

I get a strange feeling about this conversation. It's as if you have some definite positions on these matters but aren't saying what they are. It's like seeing the tip of an iceberg. I have all my cards on the table, but I don't see your hand at all. It would help if I knew what you accept and don't accept about PCT.

I accept PCT wherever it has been subject to experimental tests. I will try to stay away from extrapolations of the model to domains where data is sketchy or nonexistent. Does this reveal my hand?

Bruce

[From Rick Marken (2010.01.15.0850)]

Bruce Gregory (2010.01.15.1220 UT)--

I accept PCT wherever it has been subject to experimental tests. I will try to stay away from
extrapolations of the model to domains where data is sketchy or nonexistent. Does this reveal
my hand?

Yes. But I think of tests of PCT as tests of the overall model. Every
aspect of the overall model hasn't been tested yet and this is
definitely a problem. It's why I would like to get more researchers
involved in PCT. That way we might get some more studies done testing
the aspects of the model that have not been tested as much, such as
the model of memory. [By the way, Bill Powers has a nice little
experiment that demonstrates the use of memory in control: Powers, W.
T. (1992). A cognitive control system. In R. L. Levine & H. E.
Fitzgerald (Eds.), Analysis of dynamic psychological systems, v. 2:
Methods and applications (pp. 327�340). New York: Plenum Press. What
is missing is experimental tests of the memory processes described by
Powers in B:CP].

I agree that one should be tentative about accepting extrapolations of
the model to domains where data is sketchy or nonexistent. But I don't
think it's necessary to then accept the non-PCT based view of domains
(like memory) where data based on those views is prevalent. My feeling
is that where there is data from experiments based on the wrong
organizing principle, the data isn't worth much, even if it ostensibly
tests a domain (such as the neurophysiological basis of memory) that
has not yet been tested by experiments based on what I believe is the
correct organizing principle, PCT. That's why I personally don't have
much interest in all of the "hot" new research, neurophysiological or
otherwise, that is based on a causal model of whatever phenomenon is
under study. In his classic 1978 Psych Review paper, Bill Powers
provided a very convincing (to me, anyway) demonstration of why
behavioral science data is not of much value for understanding _any_
behavioral phenomena if it has been collected under the assumption
that the system under study is open rather than closed loop [see
Powers, W. T. (1978). Quantitative analysis of purposive systems: Some
spadework at the foundations of scientific psychology. Psychological
Review, 85, 417�435].

Best

Rick

···

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

[From Bill Powers (2010.01.15.0945 MST)]

Bruce Gregory (2010.01.15.1220 UT) --

BG: Could you be a bit more explicit about what you mean by the "empirical approach?" I think of PCT as being empirical. Does "empirical approach" function simply to identify non-control models?

As I understand the term, the empirical approach doesn't employ models at all. You simply do experiments and record the results. With statistical (or less formal) methods you can discover regularities with varying degrees of confidence, and then you can use these results to guide your actions in the future, as in diagnosis and treatment. You discover that aspirin alleviates headaches, so when a person complains of a headache, you recommend taking aspirin. No questions about what has done wrong inside when you have a headache or how headaches work. Not unless you go inside the organism and start doing chemistry there, and then there still isn't any model (other than chemistry itself).

I accept PCT wherever it has been subject to experimental tests. I will try to stay away from extrapolations of the model to domains where data is sketchy or nonexistent. Does this reveal my hand?

Closer, but I won't press. You've already helped me understand more in your statements about goals being different from reference signals. That was useful.

Best,

Bill P.

[From Fred Nickols (2010.01.15.1036 MST)]

[From Bill Powers (2010.01.15.0945 MST)]

Closer, but I won't press. You've already helped me understand more
in your statements about goals being different from reference
signals. That was useful.

Um, er. here comes another dumb question. Is there a difference between a reference "signal" and a reference "level"?

Fred Nickols
nickols@att.net

[From Rick Marken (2010.01.15.0945)]

Fred Nickols (2010.01.15.1036 MST)

Is there a difference between a reference "signal" and a reference "level"?

The reference level is the current value of the reference signal.

Best

Rick

···

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

[From Bruce Gregory (2010.01.15.1800 UT)]

[From Rick Marken (2010.01.15.0850)]

I agree that one should be tentative about accepting extrapolations of
the model to domains where data is sketchy or nonexistent. But I don’t
think it’s necessary to then accept the non-PCT based view of domains
(like memory) where data based on those views is prevalent. My feeling
is that where there is data from experiments based on the wrong
organizing principle, the data isn’t worth much, even if it ostensibly
tests a domain (such as the neurophysiological basis of memory) that
has not yet been tested by experiments based on what I believe is the
correct organizing principle, PCT. That’s why I personally don’t have
much interest in all of the “hot” new research, neurophysiological or
otherwise, that is based on a causal model of whatever phenomenon is
under study. In his classic 1978 Psych Review paper, Bill Powers
provided a very convincing (to me, anyway) demonstration of why
behavioral science data is not of much value for understanding any
behavioral phenomena if it has been collected under the assumption
that the system under study is open rather than closed loop [see
Powers, W. T. (1978). Quantitative analysis of purposive systems: Some
spadework at the foundations of scientific psychology. Psychological
Review, 85, 417–435].

Thanks, Rick. Very helpful. I’m a little puzzled as to why you are so dismissive of the neuroscience, however. A control loop, after all, consists of elements that can be studied in a strictly causal fashion. I agree that if you are not looking for feedback loops, you are unlikely to find them. Nevertheless there is so much evidence that communication in the brain goes both ways, there is plenty of grounds to think in terms of closed loops.

Bruce

[From Bill Powers (2010.01.15.1100 MST)]

Fred Nickols (2010.01.15.1036 MST) --

FN: Um, er. here comes another dumb question. Is there a difference between a reference "signal" and a reference "level"?

BP: I thought nobody would ever ask. Yes, there is, definitely. You can observe a reference level from outside the organism. You can't observe the reference signal without sticking probes into a brain.

The reference level of a variable that's under control is the value it has when there are no disturbances tending to change it. Also, it's the value toward which behavior changes the variable when it is disturbed, with the amount of action being proportional to the amount of change in the variable, and the sign of the effect of the action being opposite to the sign of the change. The reference level is an observable, measurable, property of a control loop.

The reference level can also be defined as in B:CP: it's "that level of input at which the effect of the output on the input becomes zero."

The loop gain of the control system can be calculated by dividing the amount of change in the variable (due to a known disturbance) when the control system is kept from acting, by the amount of change that occurs with the control loop intact. That ratio gives you 1 + G; subtract 1 to get the loop gain. Hope I got that right. Better work it out to see if I did.

The reference signal is an element of a proposed brain model that accounts for the observed existence of the reference level. It's hypothetical.

Best,

Bill P.

[From Bruce Gregory (2010.01.15.1835 UT)]

[From Bill Powers (2010.01.15.0945 MST)]

Bruce Gregory (2010.01.15.1220 UT) –

BG: Could you be a bit more explicit about what you mean by the “empirical approach?” I think of PCT as being empirical. Does “empirical approach” function simply to identify non-control models?

As I understand the term, the empirical approach doesn’t employ models at all. You simply do experiments and record the results. With statistical (or less formal) methods you can discover regularities with varying degrees of confidence, and then you can use these results to guide your actions in the future, as in diagnosis and treatment. You discover that aspirin alleviates headaches, so when a person complains of a headache, you recommend taking aspirin. No questions about what has done wrong inside when you have a headache or how headaches work. Not unless you go inside the organism and start doing chemistry there, and then there still isn’t any model (other than chemistry itself).

BG: O.K. I agree.

I accept PCT wherever it has been subject to experimental tests. I will try to stay away from extrapolations of the model to domains where data is sketchy or nonexistent. Does this reveal my hand?

Closer, but I won’t press. You’ve already helped me understand more in your statements about goals being different from reference signals. That was useful.

BG: I have collected a substantial amount of information on higher level behavior in conjunction with my efforts in science education. If anyone is interested, it is available on my website: www.wishfulthink.org. Critical comments are welcome. I have started to think about the extent the material might be more coherent if presented from a control theory perspective. If I come up with anything that seems to have any merit, I’ll let folks know.

Bruce

[From Bill Powers (2010.01.15.1150 MST)]

Bruce Gregory (2010.01.15.1835 UT) --

A note of appreciation. Our discussion was begining to move toward the hot side, and -- I now notice -- you turned down the heat instead of escalating any further. It works like magic, doesn't it? Thanks.

Best,

Bill P.

[From Fred Nickols (2010.01.15.1223 MST)]

[From Rick Marken (2010.01.15.0945)]

> Fred Nickols (2010.01.15.1036 MST)

> Is there a difference between a reference "signal" and a reference "level"?

The reference level is the current value of the reference signal.

That's what I thought. Thanks.

Fred Nickols
nickols@att.net

[From Rick Marken (2010.01.15.1415)]

Bruce Gregory (2010.01.15.1800 UT)

Thanks, Rick. Very helpful. I'm a little puzzled as to why you are so
dismissive of the neuroscience, however.

I didn't mean to be dismissive of neuroscience. I think neuroscience
is great. And, of course, I'm well aware of the fact that control is
implemented by the nervous system. So knowing a bit about the
"hardware" architecture (and function) is certainly useful. It's just
that much of the neuropsychology work I have seen recently does not
seem very useful. It's like the work Bill said that he saw described
at the seminar in Boulder; I think I saw the same guy give that talk
at UCLA. Bascially, what my guy did was have people do some kind of
reaction time task that involved making a choice and measuring brain
activity using an fMRI (or some high tech thing like that) while doing
it. Based on the fact that there was more activity in one area of the
brain when there was one kind of than in another when there was a
different kind of choice (I forget the details of the task) the guy
concluded that the one area of the brain was responsible for dealing
with conflicts. That strikes me as being a sophisticated,
neurophysiological phrenology. It tell me nothing about what a
conflict is, why it occurs, etc. There's no there there other than the
fancy physiological names and measures.

I guess I stopped being interested in neurophysiology when it seemed
to have become fixated more on the observational technologies than on
trying to figure out how the NS works. The neurophsiology work I
likedn best was the Hubel & Weisel receptive field stuff. That seems
very relevant to PCT. I haven't really come across any new stuff that
seems as useful.

Best

Rick

···

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

[From Bruce Gregory (2010.01.16.1302 UT)]

[From Rick Marken (2010.01.15.1415)]

I guess I stopped being interested in neurophysiology when it seemed
to have become fixated more on the observational technologies than on
trying to figure out how the NS works. The neurophsiology work I
likedn best was the Hubel & Weisel receptive field stuff. That seems
very relevant to PCT. I haven’t really come across any new stuff that
seems as useful.

I’m sympathetic. Most of what we infer about the workings of the brain come from studies of people who have had accidents, strokes, or surgery, or from animal studies where direct measurements of neural activity are possible. The high-tech imaging has at best confirmed these studies and at worst simply muddied the waters.

Bruce

[From Bruce Gregory (2010.01.16.1306)]

[From Bill Powers (2010.01.15.1150 MST)]

Bruce Gregory (2010.01.15.1835 UT) --

A note of appreciation. Our discussion was begining to move toward the hot side, and -- I now notice -- you turned down the heat instead of escalating any further. It works like magic, doesn't it? Thanks.

BG: Thanks, Bill. I appreciate the acknowledgment. I only wish the approach worked with my younger son. It seems to enrage him!

Bruce

[From Bruce Gregory (2010.01.16.1328)]

[From Bill Powers (2010.01.15.1100 MST)]

Fred Nickols (2010.01.15.1036 MST) --

FN: Um, er. here comes another dumb question. Is there a difference between a reference "signal" and a reference "level"?

BP: I thought nobody would ever ask. Yes, there is, definitely. You can observe a reference level from outside the organism. You can't observe the reference signal without sticking probes into a brain.

The reference level of a variable that's under control is the value it has when there are no disturbances tending to change it. Also, it's the value toward which behavior changes the variable when it is disturbed, with the amount of action being proportional to the amount of change in the variable, and the sign of the effect of the action being opposite to the sign of the change. The reference level is an observable, measurable, property of a control loop.

The reference level can also be defined as in B:CP: it's "that level of input at which the effect of the output on the input becomes zero."

The loop gain of the control system can be calculated by dividing the amount of change in the variable (due to a known disturbance) when the control system is kept from acting, by the amount of change that occurs with the control loop intact. That ratio gives you 1 + G; subtract 1 to get the loop gain. Hope I got that right. Better work it out to see if I did.

The reference signal is an element of a proposed brain model that accounts for the observed existence of the reference level. It's hypothetical.

BG: Nice post, Bill. Short, clear, and to the point.

Bruce