I’ve created this topic for discussion of what PCT research should look like. Perhaps the best way to start is to discuss Powers’ description of what he thought a program of research might look like. The description is in his paper “A cybernetic model for research in human development” which is reprinted in Living Control Systems I, pp. 167-219. The program is described on pp. 215-219.
Is there really no one who is interested in PCT research?
Clearly there several on this list who are.
Yes, of course. But I was hoping that some of those would who are interested in PCT research would be interested in discussing what a program of research based on PCT should look like. So I guess I should have asked if there was really no one in this group who is interested in discussing what PCT research should look like, based on Powers’ rather general ideas about this as described on pp. 167 -219 in Living Control Systems I.
Best
Rick
This part seems like a summary, from page 216, of LCSI:
I see a program of research in human development as a series of stages we go through to build a new understanding of human nature in a systematic way. We first gather basic data in a more or less naturalistic way, data which traditional methods have not provided and could not provide (*). Then we try to classify this data into a taxonomy of levels […]. Next, we can begin to look for patterns relating to the development of organisation from very little to maximum. And ultimately, I hope we can devise efficient and accurate ways for mapping the organization of an individual from his most elementary reflexes to his overall properties
A bit later, he explains that data gathering is primarily about controlled variables. Creating a catalog of controlled variables that pass rigorous tests, choosing the variables more-less from a random place in the hypothetical hierarchy. Then the variables might be classified in terms of when they appear during development, at what age; are they on the same level in the hierarchy or different ones, etc.
So, in general, the first stage of trying to understand an organism, human other animal, is to gather controlled variables. The final stage might be, as Bill talks in the preface, demonstration of understanding a system:
The best way to prove that the explanation actually explains something is to cast it as a working simulation, turn it on, and let it operate by the rules you have put in it. If you can’t do that, then you don’t have a model or an explanation.
That is a nice paragraph.
(*) Maybe the state of science today is different than 30 years ago when LCSI was written, but I think I see a lot of work with a similar goal and similar methodology. People quote Feynman’s “if you can’t build it, you don’t understand it” when discussing models of locomotion, sensory processing, brain circuit function, etc. There is the emerging field of neurorobotics aiming to connect brain science with robotic engineering, with mutual benefit. Then, various versions of control theory are used increasingly to explain locomotion. In behavior research, people might not use the expression “controlled variable”, but they do talk about “sensory cues”, “relevant stimulus features”, “non-environmental abstract variables”, and so on, often expressing the same concept.
I think a lot of controlled variables could be found in the literature, or at least some strong candidates for controlled variables. The job of a researcher might be to refine the current best guesses, verify them in experiment, and create a model, a simulation, a robot.
Anther way of doing research for someone starting from PCT is to find working models and verify that they are controlling the same variables as organisms. For example, the arm models in LCSIII have more-less arbitrary controlled variables and arbitrary arrangement in levels, not directly confirmed in experiments with human behavior. Or the movement of swarm controlled systems, etc.
E. Coli reorganization algorithms deserve a lot more attention, could be a pretty nice optimization method, I think.
RM: HI Adam. Great ideas, thanks.
AM: So, in general, the first stage of trying to understand an organism, human other animal, is to gather controlled variables.
RM: Right!
AM: I think a lot of controlled variables could be found in the literature, or at least some strong candidates for controlled variables.
RM: Yes, excellent point. A good PCT based project might be to go through the psychological, neuroscience and/ or robotics literature and make a catalog of these candidate controlled variables.
AM: The job of a researcher might be to refine the current best guesses, verify them in experiment, and create a model, a simulation, a robot.
Exactly!! And with a catalog of candidate controlled variables different researchers could do this for different entries in the catalog.
AM: Anther way of doing research for someone starting from PCT is to find working models and verify that they are controlling the same variables as organisms. For example, the arm models in LCSIII have more-less arbitrary controlled variables and arbitrary arrangement in levels, not directly confirmed in experiments with human behavior. Or the movement of swarm controlled systems, etc.
RM: Another great idea!
RM: Bill also suggested that once a sufficient number of validated controlled variables had been identified “the next step would be to sort them into meaningful categories”. You note that he mentioned that one way to do this would be to sort in terms of the age the variable becomes controllable. Since the idea that controlled variables are of different types – and indeed, are of something like eleven hierarchically arranged types, starting with intensities at the bottom and ending with system concepts at the top – can you think of another way, besides age of onset of ability to control, to test whether controlled variables might fall into classes that correspond to these types? The only one I can think of is response time.
AM: E. Coli reorganization algorithms deserve a lot more attention, could be a pretty nice optimization method, I think.
RM: I think so too. Especially given it’s practical importance in conflict resolution. Any ideas about what a study of reorganization might look like?
Best
Rick
A good PCT based project might be to go through the psychological, neuroscience and/ or robotics literature and make a catalog of these candidate controlled variables.
The last one I noticed was in balance and posture control, they were proposing the center of mass as the controlled variable and how it might be perceived trough vestibular senses and pressure in the feet, etc. Pretty interesting stuff.
Since the idea that controlled variables are of different types – and indeed, are of something like eleven hierarchically arranged types, starting with intensities at the bottom and ending with system concepts at the top – can you think of another way, besides age of onset of ability to control, to test whether controlled variables might fall into classes that correspond to these types? The only one I can think of is response time.
I think the definitions of levels are going to continue to change, as they already shifted a bit from the original ones. The levels were always a hypothesis, an instance of the principle of hierarchical control.
Response time, sure. Comparing different minimal response times might or should reflect the hierarchical relationship between control systems involved.
There is anatomy and the location of the neural parts of the control system in the brain - there are always improvements in neuroimaging, so maybe we will have tools for determining where a particular input function or comparator is located.
Maybe something like abstractness or complexity can also serve as a measure, if defined properly (I’m not too sure how). If the variable can be ‘calculated simply’, such as hand position, it might be lower in the hierarchy than something more complex, such as writing letters with a pen or typing on the keyboard. If it is single modality, it is probably lower than a multi-modal variable.
Then, there are different animals and their differences in levels. Comparing then should be interesting.
AM: E. Coli reorganization algorithms deserve a lot more attention, could be a pretty nice optimization method, I think.
RM: I think so too. Especially given it’s practical importance in conflict resolution. Any ideas about what a study of reorganization might look like?
I’m currently interested in the mathematics of the e.coli reorg. algorithm and how it compares to other gradient descent, hill-climbing, optimization algorithms, and also the so called genetic algorithms. For example, the e-coli algorithm doesn’t need to calculate the gradient of the cost function, similar to the genetic algorithms, and that might make is slower than algorithms that make use of the gradient, but maybe more robust. Not sure, really.
There is also a lot of research in the cellular mechanisms of learning in nervous systems, movement of axons during development, strengthening or weakening of synapses, etc. The literature might point to a similar learning algorithm, or the molecular basis… But… I’m not really sure what I’m talking about here, just guessing.
RM: A good PCT based project might be to go through the psychological, neuroscience and/ or robotics literature and make a catalog of these candidate controlled variables.
AM: The last one I noticed was in balance and posture control, they were proposing the center of mass as the controlled variable and how it might be perceived trough vestibular senses and pressure in the feet, etc. Pretty interesting stuff.
RM: Yes, and sometimes they even do something like a test for the controlled variable to these their hypotheses about the variable under control, for example in Lee, D. N. and Aronson, E. (1974). Visual proprioceptive control of standing in human infants. Perception & Psychophysics , 15, 529-532. But I do think a good PCT research project would be to go through the behavior science literature and catalog possible controlled variables. It could be a database with entries like:
CV Relevant Article(s) Possible Type Means of Testing
AM: I think the definitions of levels are going to continue to change, as they already shifted a bit from the original ones. The levels were always a hypothesis, an instance of the principle of hierarchical control.
RM: Yes, that’s the whole point. Bill proposed research project is all about testing the hierarchical model. In B:CP and other articles Bill presents behavioral and neurophysiological evidence (drawn from the existing literature at the time, of course) for the levels. But that was just a start. We need people do extent this work considerably. I’ve done it a little but we need a ton more.
AM: Response time, sure. Comparing different minimal response times might or should reflect the hierarchical relationship between control systems involved.
AM: There is [also] anatomy and the location of the neural parts of the control system in the brain - there are always improvements in neuroimaging, so maybe we will have tools for determining where a particular input function or comparator is located.
RM: Yes, indeed. That should all be part of the systematic collection of evidence regarding the hierarchy.
AM: Maybe something like abstractness or complexity can also serve as a measure, if defined properly (I’m not too sure how). If the variable can be ‘calculated simply’, such as hand position, it might be lower in the hierarchy than something more complex, such as writing letters with a pen or typing on the keyboard. If it is single modality, it is probably lower than a multi-modal variable.
RM: Once we see a list of descriptions of possible controlled variables we might be able come up with other ideas about how to do it. I was thinking that we might be able to have people make judgments of similarity between pairs of controlled variables (as described in the catalog) and then set some hierarchical clustering algorithms on them. Or maybe use some other type of multidimensional analysis program on the data to see if there are distinct clusters of variables in multidimensional space.
AM: Then, there are different animals and their differences in levels. Comparing then should be interesting.
RM: Yes, excellent suggestion. We could probably get a lot of possible controlled variables for animals in the comparative psychology literature.
AM: I’m currently interested in the mathematics of the e.coli reorg. algorithm and how it compares to other gradient descent, hill-climbing, optimization algorithms, and also the so called genetic algorithms…
RM: Yes, great idea. But my preference right now is for collecting data that will allow a test to see how reorganization works. Perhaps these different “learning” algorithms could be the candidate hypotheses for the kinds of learning (reorganization) that we see in real organisms.
AM: There is also a lot of research in the cellular mechanisms of learning in nervous systems, movement of axons during development, strengthening or weakening of synapses, etc. The literature might point to a similar learning algorithm, or the molecular basis… But… I’m not really sure what I’m talking about here, just guessing.
RM: About all we can do right now IS guess. These sound like very good guesses but what we can (and, I think, should) do now is collect data that allows us to test these guesses. So what I’d really like to see in this thread is suggestions for how to test guess about 1) what variables are controlled 2) how these variables are organized in a living organism and 3) how this organization develops and is maintained.
RM: `Thanks again for your contributions on this.
Best
Rick