Hierarchical network of control systems that learn...

Has anyone here seen this work before; that heavily credits Powers (see Discussion) but produce their own 'mixed' model to account for reinforcement learning?
<https://urldefense.proofpoint.com/v2/url?u=http-3A__journals.sagepub.com_doi_pdf_10.1177_105971239300100302&d=DQMCaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=-dJBNItYEMOLt6aj_KjGi2LMO_Q8QB-ZzxIZIF8DGyQ&m=x6vJNN3-a5PpO3oG202TrOMvrz77vmwwMAiDfZDDA9Y&s=usZ_1vFxNPEfEFg5A0ZpjZCTMZaZKunOja3QipRXCNA&e=>http://journals.sagepub.com/doi/pdf/10.1177/105971239300100302

Warren

···

-------- Forwarded Message --------
Subject: Hierarchical network of control systems that learn...
Date: Thu, 29 Dec 2016 22:26:14 +0000
From: Warren Mansell <mailto:wmansell@gmail.com><wmansell@gmail.com>
Reply-To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu
To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu, <mailto:CSGNET@LISTSERV.ILLINOIS.EDU>CSGNET@LISTSERV.ILLINOIS.EDU
CC: Sergio Pellis <mailto:pellis@uleth.ca><pellis@uleth.ca>, <mailto:hebell@ucsd.edu>hebell@ucsd.edu, Jeff Vancouver <mailto:vancouve@ohio.edu><vancouve@ohio.edu>, Henry Yin Ph.D. <mailto:hy43@duke.edu><hy43@duke.edu>

[From Rick Marken (2016.12.30.0945)]
I put in the header because I noticed that this was send to CSGNet as well as to others.
Hi Warren
How do you find these things?? This has to be one of the strangest PCT-based papers of all time, right up there with Carver and Scheier. I've never seen it before but I imagine Bill did. If so, I would love to have sees his response.
The paper is strange because it does heavily credit Powers (as you say) and is clear about the fact that Powers' hierarchical model is organized around control of input (perception) rather than output. But their paper is built around a model that appears to be organized around the control of output (see Figure 3 in particular). The focus of the paper is on a model of learning to it's not clear to me why they built a model of conditioning using a hierarchical control model. They are certainly not clear about what is being controlled by their model (other than "output").
So the strangeness of the paper comes from the fact that they say they are following an "agenda" pioneered by Powers, and they describe Powers' model pretty well yet they then go on to develop a model that has virtually nothing to do with Powers' model. One of their most baffling statements is that "Powers has not proposed a computational model..." (p.298). Actually, Powers presented some computational examples in the 1978 Psych Review paper to which they refer. And since their paper was published in 1993 they could have also learned about the computational implementation of Powers hierarchical model from his 1979 Byte series and well as from my 1990 "Spreadsheet Analysis" paper.
Perhaps the most surprising thing about this paper is the long quote from Ed Ford that is taken from Hershberger's "Volitional Action" book. It's quite good; well written and by and large accurate. It surprised me because I knew Ed to be a pretty poor writer. I know this because Ed wrote a little paper for the "PCT" issue of the "American Behavioral Scientist" that I edited and I had to do significant editing on his paper. So I suspect Bill Powers had a large hand in writing the quote in the Klopf paper. But if this quote is really all Ed's work then I definitely sold him short.
So Klopf et al are able to describe Powers model correctly but are unable to apply it to behavior (in this case "conditioning") correctly. I believe this results from their failure to deal with phenomena first. They might have gotten it right of they understood that the behavior involved in instrumental and classical conditioning involves control. Then they would know why they need to a control theory model to explain it -- a model that learns how to control.
Once again, the key to understanding PCT is understanding the phenomenon it explains! Phenomena first. Or, as Bill would (and did) say: Hearken to Marken;-)
Best
Rick

···

-------- Forwarded Message --------
Subject: Re: Hierarchical network of control systems that learn...
Date: Fri, 30 Dec 2016 09:45:08 -0800
From: Richard Marken <mailto:rsmarken@gmail.com><rsmarken@gmail.com>
Reply-To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu
To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu <mailto:csgnet@lists.illinois.edu><csgnet@lists.illinois.edu>
CC: Control Systems Group Network (CSGnet) <mailto:CSGNET@listserv.illinois.edu><CSGNET@listserv.illinois.edu>, Sergio Pellis <mailto:pellis@uleth.ca><pellis@uleth.ca>, Heather Bell <mailto:hebell@ucsd.edu><hebell@ucsd.edu>, Jeff Vancouver <mailto:vancouve@ohio.edu><vancouve@ohio.edu>, Henry Yin Ph.D. <mailto:hy43@duke.edu><hy43@duke.edu>

On Thu, Dec 29, 2016 at 2:26 PM, Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:

Has anyone here seen this work before; that heavily credits Powers (see Discussion) but produce their own 'mixed' model to account for reinforcement learning?
<https://urldefense.proofpoint.com/v2/url?u=http-3A__journals.sagepub.com_doi_pdf_10.1177_105971239300100302&d=DQMCaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=-dJBNItYEMOLt6aj_KjGi2LMO_Q8QB-ZzxIZIF8DGyQ&m=x6vJNN3-a5PpO3oG202TrOMvrz77vmwwMAiDfZDDA9Y&s=usZ_1vFxNPEfEFg5A0ZpjZCTMZaZKunOja3QipRXCNA&e=>> http://journals.sagepub.com/doi/pdf/10.1177/105971239300100302

Warren

--
Richard S. Marken
"The childhood of the human race is far from over. We have a long way to go before most people will understand that what they do for others is just as important to their well-being as what they do for themselves." -- William T. Powers

Thanks Rick, a very helpful critique!
Warren

···

-------- Forwarded Message --------
Subject: Re: Hierarchical network of control systems that learn...
Date: Fri, 30 Dec 2016 18:43:40 +0000
From: Warren Mansell <mailto:wmansell@gmail.com><wmansell@gmail.com>
Reply-To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu
To: <mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu
CC: Control Systems Group Network (CSGnet) <mailto:CSGNET@listserv.illinois.edu><CSGNET@listserv.illinois.edu>, Sergio Pellis <mailto:pellis@uleth.ca><pellis@uleth.ca>, Heather Bell <mailto:hebell@ucsd.edu><hebell@ucsd.edu>, Jeff Vancouver <mailto:vancouve@ohio.edu><vancouve@ohio.edu>, Henry Yin Ph.D. <mailto:hy43@duke.edu><hy43@duke.edu>

On 30 Dec 2016, at 17:45, Richard Marken <<mailto:rsmarken@gmail.com>rsmarken@gmail.com> wrote:

[From Rick Marken (2016.12.30.0945)]
I put in the header because I noticed that this was send to CSGNet as well as to others.
Hi Warren
How do you find these things?? This has to be one of the strangest PCT-based papers of all time, right up there with Carver and Scheier. I've never seen it before but I imagine Bill did. If so, I would love to have sees his response.
The paper is strange because it does heavily credit Powers (as you say) and is clear about the fact that Powers' hierarchical model is orga nized around control of input (perception) rather than output. But their paper is built around a model that appears to be organized around the control of output (see Figure 3 in particular). The focus of the paper is on a model of learning to it's not clear to me why they built a model of conditioning using a hierarchical control model. They are certainly not clear about what is being controlled by their model (other than "output").
So the strangeness of the paper comes from the fact that they say they are following an "agenda" pioneered by Powers, and they describe Powers' model pretty well yet they then go on to develop a model that has virtually nothing to do with Powers' model. One of their most baffling statements is that "Powers has not proposed a computational model..." (p.298). Actually, Powers presented some computational examples in the 1978 Psych Review paper to which they refer. And since their paper was published in 1993 they could have also learned about the computational implementation of Powers hierarchical model from his 1979 Byte series and well as from my 1990 "Spreadsheet Analysis" paper.
Perhaps the most surprising thing about this paper is the long quote from Ed Ford that is taken from Hershberger's "Volitional Action" book. It's quite good; well written and by and large accurate. It surprised me because I knew Ed to be a pretty poor writer. I know this because Ed wrote a little paper for the "PCT" issue of the "American Behavioral Scientist" that I edited and I had to do significant editing on his paper. So I suspect Bill Powers had a large hand in writing the quote in the Klopf paper. But if this quote is really all Ed's work then I definitely sold him short.
So Klopf et al are able to describe Powers model correctly but are unable to apply it to behavior (in this case "conditioning") correctly. I believe this results from t heir failure to deal with phenomena first. They might have gotten it right of they understood that the behavior involved in instrumental and classical conditioning involves control. Then they would know why they need to a control theory model to explain it -- a model that learns how to control.
Once again, the key to understanding PCT is understanding the phenomenon it explains! Phenomena first. Or, as Bill would (and did) say: Hearken to Marken;-)
Best
Rick

On Thu, Dec 29, 2016 at 2:26 PM, Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:

Has anyone here seen this work before; that heav ily credits Powers (see Discussion) but produce their own 'mixed' model to account for reinforcement learning?
<https://urldefense.proofpoint.com/v2/url?u=http-3A__journals.sagepub.com_doi_pdf_10.1177_105971239300100302&d=DQMCaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=-dJBNItYEMOLt6aj_KjGi2LMO_Q8QB-ZzxIZIF8DGyQ&m=x6vJNN3-a5PpO3oG202TrOMvrz77vmwwMAiDfZDDA9Y&s=usZ_1vFxNPEfEFg5A0ZpjZCTMZaZKunOja3QipRXCNA&e=>>> http://journals.sagepub.com/doi/pdf/10.1177/105971239300100302

Warren

--
< div dir="ltr">

Richard S. Marken
"The childhood of the human race is far from over. We have a long way to go before most people will understand that what they do for others is just as important to their well-being as what they do for themselves." -- William T. Powers