The Study of Living Control Systems: A Guide to Doing Research on Purpose 9781108485586, 9781108752138, 2020036902, 2020036903, 9781108707336

Citation preview

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

T H E ST U DY OF L I V I NG CON T ROL S YST E MS

This book is a guide to doing a new kind of psychological research that focuses on the purposes rather than the causes of behavior. The research methods described here are based on a theory of behavior called Perceptual Control Theory (PCT) that views organisms as purposeful rather than mechanistic systems. According to PCT, purposeful behavior involves acting to control perceptual input variables. Thus, understanding the purposeful behavior of living organisms is a matter of determining the perceptual variables they are controlling when they are carrying out various behaviors. This book outlines research methods that determine what perceptual variables an organism is controlling, how it controls those variables, and why. It also describes methods for studying how an organism develops the ability to control different perceptions and how consciousness might be involved in this process. r ich a r d s. m a r k e n is a leading authority on the Perceptual Control Theory (PCT) model of behavior. Now retired from a career as a psychology professor, human factors engineer, and health policy researcher, he still actively consults and carries out a program of research testing PCT.

Endorsements for Living Control Systems The history of science is marked by revolutions that are advanced by novel methods of observation and experiment. Richard Marken provides a comprehensive and indispensable research guide to a scientific revolution still in the making: understanding the purposeful nature of the behavior of living organisms as they act as living control systems. Gary Cziko, Professor Emeritus of Educational Psychology, University of Illinois at Urbana-Champaign, USA This book provides, with practical examples, some much-needed insight into how to study what living things do from beyond a stimulus–response perspective. This understanding has wide-ranging consequences for the study of behavior. Heather Broccard-Bell, Adjunct Assistant Professor in Psychological Sciences, University of San Diego, USA This is a beautifully crafted book that provides a refreshingly different perspective on research. Each chapter is like opening a door into a whole new way of thinking about what we already thought we knew. This book is a must for both novice and experienced researchers. Sara Tai, Senior Lecturer in Clinical Psychology, University of Manchester, UK Behavior serves a purpose, instead of being a reaction to a stimulus. What does this imply for a new science of psychology? What kind of theorizing, modeling, experimentation is adequate? Seriously occupied with these questions, I read this book. I was blown away by the creative, often surprising, insights and advice. Franz Mechsner, Associate Professor, Northumbria University, UK Richard Marken, one of the finest experimental psychologists of our time, has written a concise and readable introduction to Perceptual Control Theory. It will be a valuable resource to all students studying behavior. Henry Yin, Professor of Psychology and Neuroscience, Duke University, USA This book successfully turns the spotlight on experimental methodology for testing living control systems. Richard Marken describes methods based on Perceptual Control Theory (PCT) that can be used to study learning, social, cognitive, and psychotherapeutic constructs. Both novice and expert researchers should read this survey of PCT research. Grace B. Dyrud, Emeritus Professor of Psychology, Augsburg University, USA

Richard Marken leads the academic world in the study of living control systems. The research he describes takes forward pioneering methodologies. Full of diverse examples and illustrative diagrams, this book now makes these transformative methods practical for researchers, practitioners, and students across the life and social sciences. Warren Mansell, Reader in Clinical Psychology, University of Manchester, UK, and Editor of The Interdisciplinary Handbook of Perceptual Control Theory: Living Control Systems IV In this book, Richard Marken provides researchers with the information necessary to design and re-examine research in the field of psychology. It is a must-read for anyone who wants to do quality research in the fields of cognition, experimental psychology, consciousness, and behavior. Shelley A. W. Roy, Senior Faculty Member for the International Association of Applied Control Theory

THE STUDY OF LIVING C O N T RO L S Y S T E M S A Guide to Doing Research on Purpose Ri chard S. Ma rk e n University of California–Los Angeles

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108485586 DOI: 10.1017/9781108752138 © Richard S. Marken 2021 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2021 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Marken, Richard S., 1946- author. Title: The study of living control systems : a guide to doing research on purpose / Richard S. Marken, University of California, Los Angeles. Description: New York, NY : Cambridge University Press, 2021. | Includes bibliographical references and index. Identifiers: LCCN 2020036902 (print) | LCCN 2020036903 (ebook) | ISBN 9781108485586 (hardback) | ISBN 9781108707336 (paperback) | ISBN 9781108752138 (ebook) Subjects: LCSH: Psychology--Research. | Perceptual control theory. | Human behavior. Classification: LCC BF76.5 .M335 2021 (print) | LCC BF76.5 (ebook) | DDC 150.72/1--dc23 LC record available at https://lccn.loc.gov/2020036902 LC ebook record available at https://lccn.loc.gov/2020036903 ISBN 978-1-108-48558-6 Hardback ISBN 978-1-108-70733-6 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Figures

page viii

List of Tables

x

Preface

xi

1

Living Control Systems

1

2

Doing Research on Purpose

18

3

Getting Started

35

4

Basic Research on Purpose

48

5

Exploring the Hierarchy

68

6

Learning

89

7

Social Control

104

8

Back to the Future (of PCT Research)

122

Bibliography

131

Index

137

vii

Figures

1.1 Control theory model of a control system (after Powers, 1973a, Figure 1) page 3 1.2 Manual Control Theory (MCT) mapping of control theory to behavior 8 1.3 One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2) 12 1.4 An insect avoidance task that illustrates the behavioral illusion 14 2.1 One frame of the animated computer display in the Mind Reading demo 25 2.2 (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2) 30 3.1 Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2) 44 4.1 Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4) 49 4.2 Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5) 51 4.3 Multiple choices of controlled variables (Powers, 1978, Figure 8) 54 4.4 Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1) 59 4.5 Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment 62 viii

Endorsements for Living Control Systems 4.6 Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder 4.7 Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5) 5.1 Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3) 5.2 Setup of the coordination experiment 5.3 Two-level model of coordinated movement 5.4 Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3) 5.5 Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4) 6.1 The reorganization system in relationship to the hierarchy of control systems 6.2 Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1) 7.1 Tracks to the final state of the “guru” scenario of the CROWD simulation 7.2 Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation

ix 64 65 69 74 75

84 85 90 94 107 110

Tables

5.1 Possible levels in a hierarchy of control systems 8.1 A portion of a possible catalog of controlled variables

x

page 77 124

Preface

This book is a guide to doing a new kind of psychological research that is aimed at understanding the purposes rather than the causes of behavior. If you have already taken a course on research methods in psychology you will see that the methods described here differ considerably from the ones described in those courses. Indeed, the difference is apparently so great that, to date, these methods have lingered somewhat outside the mainstream of psychological research. But I believe the reason for this has more to do with pragmatism than novelty. Mainstream researchers don’t ignore new methodologies simply because they are different. If they did then a journal such as Psychological Methods, which introduces new methodologies in every issue, would have a much lower impact factor than it has. Rather, researchers ignore new methods that seem unnecessary; that can’t help them achieve their goals. Since the goal of most psychological research is to understand the causes of behavior, it is not surprising that researchers would see methods aimed at understanding the purposes of behavior as being unnecessary. So my aim in this book is not only to describe a new approach to doing psychological research but also to explain why this new approach is absolutely necessary. The book starts by explaining that the research approach described here is necessary because it is the only way to find out how the behavior of a living control system actually works. A living control system is a system that controls in the sense that it acts to keep aspects of its own experience – its perceptions – in preselected states, protected from the effects of disturbances that would move them from these states. That is, the behavior of a living control system can be described as the control of perception (Powers, 1973b). This is purposeful behavior – the purpose being to keep perceptions in preselected states. In order to understand such behavior one has to know what perceptions the system is controlling. The research methods described here are aimed at doing just that: determining xi

xii

Preface

the perceptual variables that a living control system is controlling when it is seen carrying out various behaviors. If organisms are living control systems – and there is considerable evidence that they are – then the methods described in this book are the only ones that are appropriate to the study of their behavior. This is because they are the only methods that can reveal the perceptual variables around which their behavior is organized – what are called controlled variables. The conventional methods of psychological research completely ignore the existence of controlled variables. Instead, their aim is to find evidence of causal relationships between independent (environmental) variables and dependent (behavioral) variables. But there is reason to believe that these relationships tell us little about the nature of the organisms under study. Perceptual Control Theory (PCT) – a theory that explains how living control systems work – shows us that, if organisms are living control systems, then the independent–dependent variable relationships that are found in conventional psychological research are actually side effects of the disturbance-resisting nature of these systems and, therefore, tell us more about the nature of the environment in which these systems do their behaving than about the systems themselves (Powers, 1978). Thus, the pragmatic reason for doing psychological research using the methods described here is provided by PCT. The reason is that these are the only methods that provide a correct picture of the nature of the organisms under study – if those organisms are living control systems. The reasons for thinking that organisms are, indeed, living control systems are presented in Chapter 1, where we see that the purposeful behavior of organisms is equivalent to the controlling done by nonliving control systems, such as a thermostat. PCT is, therefore, an explanation of how both nonliving and living control systems work. The research methods described in the remainder of the book show how to test different predictions of the PCT model of behavior. These different predictions are derived from the complete version of PCT, which views organisms as a hierarchy of control systems, where systems at each level of the hierarchy are controlling different types of perceptual variables; higher level systems control more complex perceptions than lower level ones. This aspect of PCT is meant to account for the fact that organisms carry out purposes of different levels of complexity; carrying out the purpose of pointing a finger, for example, involves control of a less complex perception, the perception of the position of the finger, than carrying out the purpose of making a point in a political debate, which involves control of a far more

Preface

xiii

complex perception – the perception of one’s position on a political issue. This hierarchical model leads to predictions about the types of perceptual variables that organisms control and how they control them – predictions that can only be tested using the methods described in the book. In order to test any of the predictions of PCT, the researcher must be able to determine what variable or variables the organism is controlling when it is carrying out various behaviors. This is done using the test for the controlled variable or TCV. The TCV is both the centerpiece and the most misunderstood aspect of the approach to research described here. It is the centerpiece of this approach because it is aimed at determining the variables around which behavior is organized; and when you know what these variables are, you know nearly everything you need to know about why the organism does what it does. But the TCV is also the most misunderstood aspect of this approach to research because it is often taken to be a test to determine the control variable, a variable that controls the behavior of the organism, rather than the controlled variable – a variable that the organism controls. This misunderstanding seems to come from a desire to see the TCV as being compatible with the conventional approach to psychological research where the goal is to find the variables that control behavior. By simply dropping the “ed” the controlled variable becomes an independent variable – a variable that is presumed to control (meaning to cause) behavior – and the TCV can be seen as a version of the conventional approach to doing psychological research, which it definitely is not. One of my main reasons for writing this book is to put the “ed” back into the “controlled variable.” Controlled variables are discussed throughout this book; control variables don’t show up at all. By putting “ed” back where it belongs I hope to keep the reader aware of the fact that the study of living control systems is aimed at determining the variables that are controlled by – not the variables that control – these systems. This book is based on the work of the late William T. Powers, who developed the PCT model of purposeful behavior and described the methods to test it that are described herein. I wish I had been able to write this book while Bill was still with us; besides being a brilliant scientist he was a skillful teacher and patient critic. But as it is, I managed to get useful advice during the writing of this book from a number of very capable colleagues and friends including, in alphabetical order, Professor Heather Bell, Professor Tim Carey, Professor Grace Dyrud, Professor Warren Mansell, Mr. Jeff Rothenberg, and Ms. Rikki Westerschulte. I would like

xiv

Preface

to express particular gratitude to Dr. Ryan Hughes for help with the section on neurophysiological evidence for levels of control and to Professor Henry Yin for encouraging me to write this book in the first place. I hope that the result justifies Henry’s confidence in me.

1

Living Control Systems

The study of living control systems is the study of the behavior of living organisms. Organisms can be regarded as living control systems because their behavior is equivalent to that of nonliving control systems. The equivalence turns on the fact that both nonliving control systems and living organisms control, which means that they act to achieve intended or goal results in the face of unpredictable, and often undetectable, disturbances that would prevent these results from being achieved. For example, the thermostat – a nonliving control system – controls by acting to keep room temperature constant in the face of disturbances, such as variations in the number of people in the room, that would otherwise cause the temperature to vary considerably. Similarly, a person sipping tea controls by acting to get the cup consistently to their lips in the face of disturbances, such as the changing weight of the cup after each sip, that would otherwise cause “many a slip between cup and lip.”1 The research methods described here are based on the fact that the behavior of organisms is a process of control. We know this because everything we see organisms doing – all of their behavior – is done in a world of continuously varying disturbances. These disturbances should make it impossible for organisms to produce the consistent results that we call their behaviors. Instead, we see organisms producing consistent results in the face of these disturbances. We see people consistently lifting cups to their lips without spilling a drop, a behavior called “sipping tea”; we see people consistently putting one foot in front of the other without falling, a behavior called “walking.” The study of the behavior of living control systems is, therefore, the study of how organisms produce consistent results in the face of disturbance; that is, it is the study of how organisms control. But it could also be called the study of how organisms carry out their 1

A video demonstration of the controlling involved in sipping tea is available at www.youtube.com/ watch?v=88aXMEgvq68.

1

2

Living Control Systems

purposes because the controlling done by living organisms is equivalent to what is called purposeful behavior.

1.1  Purposeful Behavior The behavior of control systems, like that of living organisms, is goaloriented. This was enough to convince some scientists that the behavior of a control system is purposeful and, thus, could be used as a scientific basis for distinguishing purposeful from the nonpurposeful behavior (e.g., Rosenblueth, Wiener, & Bigelow, 1943). These scientists were right about the behavior of a control system being purposeful but wrong about goal-orientation being enough to distinguish purposeful from nonpurposeful behavior. The fact that goal-orientation alone is not enough to distinguish purposeful from nonpurposeful behavior was pointed out by the pioneering psychologist William James in his parable of Romeo and the iron filings (James, 1890). In that parable James notes that the behavior of iron filings moving to a magnet appears to be as goal-oriented as the behavior of Romeo moving to Juliet. Yet the iron filings have no purpose while Romeo certainly does. And the way to show this is by placing an obstacle in the paths to their goals; a card can be placed between the filings and the magnet and a wall between Romeo and Juliet. What we will see is that the filings are stopped by the card and never get to their “goal,” while Romeo does whatever is necessary to get past the wall and close to Juliet. As James puts it, “With the filings the path is fixed; whether it reaches the end depends on accidents. With the lover it is the end which is fixed, the path may be modified indefinitely” (James, 1890, p. 7). James correctly understood purposeful behavior to be a process of achieving goals (“ends”) by varying actions as necessary (“modifying the path indefinitely”) in order to compensate for disturbances (“obstacles”) that would prevent goal-achievement. But he didn’t understand how such behavior was possible since it seems to violate the law of cause and effect – that cause must precede (or, at least, be simultaneous with) effect. In purposeful behavior, this temporal relationship between cause and effect seems to be reversed: a future event – the goal – seems to be the cause of the present actions that are used to achieve it. After James described purposeful behavior, many efforts were made to explain it without violating the law of cause and effect. This was done by either ignoring the fact that organisms achieve their goals in the face of disturbances (e.g., Turvey, Shaw, & Mace, 1978) or by assuming that these disturbances had the remarkable ability

1.2  Control Theory

3

to cause just the right actions that would get organisms past them and to their goals (Tolman, 1922).

1.2  Control Theory It is now possible to explain purposeful behavior without denying the reality of disturbances or attributing an unlikely level of intelligence to them. And it can be done in a way that is perfectly consistent with the law of cause and effect. It is done by recognizing that the purposeful behavior of organisms is equivalent to the controlling done by a control system. The way control systems do this controlling is explained by an engineering model called control theory.2 The control theory model of a control system is shown in Figure 1.1. The first thing to notice about this model is that it distinguishes the system doing the controlling from the environment in which this controlling is done. The dashed line in the figure encloses the control system (the System), which can be thought of as being equivalent to the organism doing the controlling. Everything outside of the dashed line is the Environment in which the system does its controlling. If the System is thought of as an organism then the Environment is the world outside of the organism’s nervous system, which includes its muscles and

r(t)

C p(t)

e(t)

I

System Environment

d(t)

v

v v

O qo(t)

F

qi(t) Controlled Variable

Figure 1.1  Control theory model of a control system (after Powers, 1973a, Figure 1). 2

A very good nontechnical introduction to engineering control theory can be found in A History of Control Theory by Bennett (1993).

4

Living Control Systems

glands. The most important variable in the System’s Environment is the controlled variable, symbolized qi(t), which is the variable that the system is controlling.3 The controlled variable is typically a function of many physical variables, represented by the v’s in the diagram. For example, Romeo is controlling his distance from Juliet. The distance from Romeo to Juliet is a controlled variable that is a function of two physical variables, which are the geographical locations of Romeo and Juliet. A controlled variable is controlled in the sense that it is kept in a goal or reference state, protected from the effects of disturbances, d(t). The reference state of the controlled variable is specified by a reference signal, r(t), inside the control system. The reference state specified by r(t) for the variable Romeo is controlling – the distance between him and Juliet – is “close to” or “near.” A control system brings the controlled variable to the reference state and keeps it there – it achieves its purpose – by varying its actions, qo(t), in exactly the right way so as to oppose the effects of disturbances. Romeo controls for being close to Juliet by acting in just the right way, by scaling walls and dodging through forests, so as to oppose the disturbances created by walls and Capulets. The System side of the control theory model explains how this is done. That is it explains how the System is able to do exactly what is required in order to achieve the purpose of keeping the controlled variable in the reference state. An important thing to notice about the System component of Figure 1.1 is that it is in a closed-loop relationship with respect to the controlled variable. This can be seen in the ring of arrows going into the System from the controlled variable, looping through the System, coming out and looping back through the Environment to where the loop started, at the controlled variable. A closed loop like this has no beginning or end. But when we describe the loop we have to start somewhere, and the typical place to start is with the input to the System – the controlled variable. The physical variables that are the basis of the controlled variable – the v’s in Figure 1.1 – enter the System via an Input Function, I, that produces a perceptual signal, p(t), that is an analog of the controlled variable. The perceptual signal then enters a comparator function, C, where it is continuously compared to a reference signal, r(t), which specifies the desired state or value of p(t). The result is an error signal, e(t), which is proportional to the difference between the reference and perceptual signals. This error signal enters an output function, O, that produces an output, qo(t), which is an action that 3

The “t” in parentheses next to the variable name (qi in this case) simply means that the variable can vary over time.

1.3  Perceptual Control Theory (PCT)

5

is proportional to the size of the error signal. This output action completes the closed loop by “feeding back” through the environment, via a feedback function, F, to have an effect on the controlled variable, which was where we started. A System that is in a closed-loop relationship with its Environment is a control system only if the feedback from its output to its input is negative. There is negative feedback in a closed loop when the error signal causes outputs that reduce – that is, have a negative effect on – the error signal itself. The error signal does this by continuously causing outputs that counter the causes of error. The causes of error are the effects of disturbances to the controlled variable as well as changes in the value of the reference signal, r(t), that specifies the desired state of the perceptual signal. This negative feedback organization will keep the controlled variable in a reference state that corresponds to the state of the perceptual signal specified by the possibly varying reference signal. The behavior of the control system model of purposeful behavior shown in Figure 1.1 can be described by the following two equations:

qo (t ) = O ⋅ [r (t ) − p(t )],

(1.1)



p(t ) = qi (t ) = I ⋅ [ F ⋅ qo (t ) + d (t )].

(1.2)

In order to simplify the mathematics, the functions in these equations – the output function, O, and the input function, I ,– are shown as constant multipliers. Equation (1.1) is called the system equation because it describes the input–output characteristics of the control system: the output of the system, qo(t), is proportional to the error signal, e(t), which is the difference between the perceptual signal, p(t), and the reference signal, r(t); r(t) p(t). Equation (1.2) is called the environment equation because it describes how variables in the environment part of the loop, including the output of the control system itself, affect the perceptual input to the system, p(t). Since p(t) is an analog of the controlled variable, qi(t), its value is ultimately determined by the combined effects of system output, qo(t), and disturbances, d(t).

1.3  Perceptual Control Theory (PCT) Equations (1.1) and (1.2) describe the causal relationships between variables that exist in a closed negative feedback loop. This shows up in the fact that system output, qo(t), is a function of system input, p(t), per Eq. (1.1), while

6

Living Control Systems

system input is a function of system output, per Eq. (1.2). Moreover, the relationships between system input and output described by Eqs. (1.1) and (1.2) are happening at the same time; input is causing output, while output is causing input. Therefore, in order to understand what this control system is doing in terms of the behavior of its output and input, Eqs. (1.1) and (1.2) must be solved simultaneously. When we do this, making the appropriate assumptions about the values of I, O, and F, we get:

qo (t ) ≈ r (t ) − 1 / F ⋅ d (t ),

(1.3)



p(t ) ≈ r (t ).

(1.4)

These equations describe the two main characteristics of the behavior of a properly functioning control system: disturbance resistance and perceptual control. Equation (1.3) describes the disturbance resistance characteristic of control system behavior. This can be seen most clearly if we make r(t) a constant equal to 0. In that case, Eq. (1.3) becomes qo(t) ≈ −1/F·d(t), which says that variations in the output of the control system are negatively related to variations in disturbances to the controlled variable. This means that system outputs compensate for or “resist” the effects that disturbances would otherwise have on the controlled variable. Equation (1.4) describes the perceptual control characteristic of control. It says that the control system keeps its perceptual signal, p(t), approximately equal to the reference signal, r(t). The reference signal functions as a specification for the “goal” state of the perceptual signal. The perceptual signal is controlled in the sense that it is brought to this goal state and maintained there in the face of disturbances. Since the perceptual signal is an analog of the controlled variable, Eq. (1.4) means that a control system will keep the controlled variable, qi(t), in a reference state, qi(t)*, which corresponds to the value specified by the reference signal. More succinctly, Eq. (1.4) says that the behavior of a control system is the control of perception in the sense that the system acts to keep a perceptual analog of the controlled variable in a reference or goal state (Powers, 1973b, 2005b). This fact about control system operation is particularly important when control theory is applied to the behavior of living organisms. This is because understanding the behavior of living organisms in terms of control theory is largely a matter of trying to figure out what perceptual variables they control (Marken, 2020). So when control theory is applied to the behavior of living organisms we give it a special name: Perceptual Control Theory or PCT.

1.4  Perceptual versus Manual Control Theory

7

Equations (1.3) and (1.4) describe the behavior of a properly functioning control system. The main factor that affects how well a control system controls is the relationship between its loop gain and speed of response. Loop gain is a measure of the “strength” of the control system; it is mainly determined by the degree to which the output function, O, amplifies the effect of error on the controlled variable. Speed of response depends on how quickly error is turned into the output that affects the controlled variable. A control system will function properly only if loop gain and speed of response are inversely related; the greater the system’s loop gain, the slower must be the system’s speed of response. This relationship between loop gain and speed of response exists in any properly functioning living control system. Since our main concern in this chapter will be the behavior of living control systems that are functioning properly – that are “in control” – we can assume that there is the correct inverse relationship between loop gain and speed of response in these systems. This means that, when studying the behavior of living control systems, the requirement that there be an inverse relationship between loop gain and speed of response can be safely ignored. However, when we build models of the controlling done by living systems, we will have to take this relationship into consideration in order to make the models “work.”

1.4  Perceptual versus Manual Control Theory PCT is not the only application of control theory to understanding the behavior of organisms. Another approach, which started just after World War II, was also aimed at evaluating human performance in manual control tasks, such as flying airplanes (Craik, 1947, 1948). Because of the emphasis on the study of manual control, this approach to the application of control theory can be called Manual Control Theory or MCT. Both PCT and MCT use the same control theory to model behavior. The difference is in the way control theory is mapped to the behaving system, which results from a difference in the way behavior is viewed. MCT views behavior as output caused by stimulus input, whereas PCT views behavior as the control of input. The PCT view of behavior results in the mapping shown in Figure 1.1; the MCT view of behavior results in the mapping shown in Figure 1.2. Figure 1.2 is the typical control system diagram that is found in texts on the MCT approach to understanding human behavior. The names of the variables and functions in Figure 1.2 have been selected so that they correspond to those in Figure 1.1. The System in Figure 1.2 represents an organism as an input–output device. Input is converted into output by a

8

Living Control Systems System

r(t)

e(t)

O

qo(t)

Plant

F

qi(t)

Output

d(t)

Figure 1.2  Manual Control Theory (MCT) mapping of control theory to behavior.

transfer function, O, which corresponds to the output function in Figure 1.1. As in Figure 1.1, the output function converts the error signal, e(t), into the output variable, qo(t). But in this case, the error signal is outside the boundary of the System (the dashed lines), in the environment. The error signal is produced by a comparator (the circle around an “X”) that subtracts the fed-back “Output” of the System – the variable called qi(t) in Figure 1.1 – from a reference signal, r(t). System output, qo(t), is converted into the fed-back Output, qi(t), by the feedback function, F, as it is in Figure 1.1. The MCT mapping of control theory to behavior is designed to be consistent with an input–output or stimulus-response view of behavior. The behaving System is viewed as a transfer function (the function O) that converts stimulus input, e(t), into behavioral output , qo(t). MCT uses control theory to evaluate characteristics of this transfer function, such as the speed of response to stimulus input or the effect of the spectral composition of disturbances on the conversion of input into output (Sheridan & Ferrell, 1974; Jagacinski & Flach, 2002; Wickens et al., 2012). The most obvious difference between the MCT and PCT mappings of control theory onto behavior is in the location of the reference signal, r(t). In control theory, r(t) specifies the result to be produced by the control system. According to MCT, organisms produce behavioral results that are “specified” by events in the environment, so r(t) is placed in the System’s environment. According to PCT, organisms produce perceptual results that are specified by events inside the organism itself, so r(t) is placed inside the System. This apparently small difference in the way control theory is mapped to behavior leads to a significant difference in the goals of research aimed at understanding the behavior of organisms. Research based on MCT is aimed at understanding characteristics of the organism’s output (or transfer) function, O. PCT research, on the other hand, is aimed at understanding characteristics of the organism’s perceptual (or input)

1.5  Controlled and Perceptual Variables

9

function, I. This difference in research objectives is highlighted by the fact that the input function, I, is a prominent feature of the PCT diagram in Figure 1.1, but is not to be found in the MCT diagram in Figure 1.2. The MCT mapping of control theory onto behavior makes the mistake of placing the reference signal and, thus, the error signal, in the environment. The error signal, which drives the output of a control system, represents a discrepancy between what an organism wants, r(t), and what it is getting, qi(t). A control theory model of organisms should place r(t) – and e(t) – inside the System. Placing r(t) in the System’s environment is an example of what Powers (1978) called the man-machine blunder. The effect of this blunder on behavioral research was described by Powers as follows: If one’s primary purpose is to keep pilots from flying airplanes into the ground or to make sure that a gunner hits a target with the shell, that is, if one’s purposes concern objectivized side effects of control behavior, the man-machine blunder amounts to nothing worse than a few mislabelings having no practical consequences. If one’s interest is in the properties of persons, however, the man-machine blunder pulls a red herring across the path of progress. (Powers, 1978, p. 419)

An objectivized side effect of control is what happens in the environment while a person (or other organism) is controlling its own perceptions. Clearly, there are many practical situations where knowing about these side effects is extremely important. For example, it’s important to know what is happening to the actual attitude of an airplane while the pilot is controlling her perceptions of the displayed values of pitch, roll, and yaw in the cockpit. It is possible to learn how to prevent unwanted side effects of control – especially poor control – without learning too much about how a person does this controlling. But if you want to understand how organisms “work” – how their internal properties function to make it possible for the organism to control various aspects of the environment – you have to do research aimed at determining the perceptual variables being controlled when they are carrying out various behaviors. When you do this kind of research – research on purpose – you will learn not only how organisms control but also what the objectivized side effects of this controlling will be when they don’t control well.

1.5  Controlled and Perceptual Variables According to PCT the variables organisms control are perceptual variables constructed from the sensory effects of environment variables. These

10

Living Control Systems

environmental variables are the v’s inside the circle representing the controlled variable, qi(t), in Figure 1.1. The “construction” is done by the input function, I, of the control system. The result of this “construction” is presumed to be an afferent neural signal that fires at a rate proportional to the value of the controlled variable – the variable constructed by the input function. The variables constructed by input functions do not necessarily correspond to a “real” entity in the environment. This means that controlled variables are perceptions themselves in the sense that they are aspects of the environment – functions of the v’s in Figure 1.1 – that can be perceived by an observer of the behaving system. For example, the taste of lemonade is the state of a perceptual variable, p(t), that is constructed from the sensory effects of sugar (v1), acid (v2), and oils (v3) mixed with water (v4).4 A simplified version of the input function that constructs this taste variable might look like this:

p(t ) = k0 + k1 * v1 (t ) + k2 * v2 (t ) + k3 * v3 (t ) + k4 * v4 (t ),

(1.5)

where p(t) is a taste perception created by the input function that is a linear combination of the environmental variables, vi(t). The ki are the coefficients that define how these variables combine to produce the taste perception. When the appropriate amounts of sugar (v1), acid (v2), oils (v3), and water (v4) are mixed together, the taste perception is that of lemonade. As Powers notes, however, no matter how real this perception seems, “there is no physical entity that corresponds to it” (Powers, 2005b, p. 112). Since perceptual variables are assumed to be analogs of controlled variables, the same function that defines the perceptual variable, p(t), also defines the controlled variable, qi(t). It is in this sense that both qi(t) and p(t) are perceptual variables. Both are functions of the sensory effects of physical variables. While the functions that produce p(t) are the input functions, I, in the organism under study, the functions that produce qi(t) are the input functions of the human who is observing the organism’s behavior. The input functions of the observer can be the person’s perceptual functions, but more often these functions are carried out by human-made devices that allow the observer to perceive aspects of the environment that could not be perceived without them, such as the otherwise undetectable acoustical echoes controlled by bats.

4

This example is based on one described by Powers (2005b, p. 112).

1.6  Confusions and Illusions

11

The fact that an observer can use a human-made device to perceive what the system under study is perceiving means that it is not necessary to know how that system’s input functions construct the perceptual variables it controls. An observer can use these human-made devices to perceive the same perceptual variables the system under study is controlling without having to know how that the system’s nervous system constructs those variables. The dB meter used to perceive the intensity of very high frequency sounds, for example, constructs this perception in a way that is quite different than the way the bat does it. Nevertheless, there is considerable evidence that the reading provided by that meter corresponds to the perception the bat controls (Griffin, 1974).

1.6  Confusions and Illusions The two characteristics of control described by Eqs. (1.3) and (1.4) – disturbance resistance and control of perception – have important implications for how to study living control systems. The disturbance resistance described by Eq. (1.3) tells us that the behavior of a control system can be mistaken for that of a stimulus-response device (Powers, 1978). This could be called the “S-R error” and we are making it when we see a disturbance, such as a tap on the patellar tendon, as a stimulus that causes a response, such as the observed leg kick or “knee jerk.” The S-R error occurs when we fail to see that the apparent response to a stimulus is the action of a control system as it resists a disturbance to a controlled variable. In the case of the knee jerk reflex, the controlled variable is the degree of stretch of the quadriceps muscle. The tap on the tendon disturbs this variable by stretching the quadriceps. The response that resists this disturbance is contraction of the quadriceps and relaxation of the hamstring muscles, a response that is aimed at bringing the degree of stretch of the quadriceps back to the reference level. The leg kick results from the fact that this response is resisting a disturbance – the stretch produced by the tap – that is no longer present. 1.6.1  Confusions The S-R error has important implications for the study of living control systems because the S-R appearance of control can blind one to the existence of controlled variables, especially when these S-R relationships are particularly dramatic. This phenomenon, which has been dubbed “control blindness,” was demonstrated in a study performed by Willett et al. (2017).

12

Living Control Systems

Figure 1.3  One frame of a video of the rubber band demo showing P’s hand movements (left side) made in “response” to E’s hand movements as the means of keeping the knot linking the rubber bands over a target dot (adapted from Willett et al., 2017, Figure 2).

Participants in the study were asked to watch a video of a demonstration of control behavior called the rubber band demo (Powers, 1973b, 2005b). One frame of the video is shown in Figure 1.3. The hand in the lower right of the frame is that of an experimenter, E; the hand in the upper left of the frame is that of the participant, P. Both hands hold pens that are attached to opposite ends of two rubber bands that are knotted together. E has instructed P to keep the knot in the rubber bands above a target dot on the paper below. (The target dot can be seen just to the right of the knot in the rubber bands.) E then disturbs the position of the knot by moving her pen in an arbitrary pattern that is traced out by the squiggle in the lower right of the figure. In order to keep the knot over the target dot, P must precisely oppose these disturbances by making appropriate movements of his pen. These opposing movements are shown as the squiggle in the upper left of the figure. P’s movements trace out a rough mirror image of E’s movements because, in order to keep the knot over the dot, P’s movements of the rubber band must be the opposite of E’s movements. The distance between the knot and the dot is the variable that P is controlling; the controlled variable, qi(t). The reference state of this variable is zero distance between the knot and dot. E’s movements are a disturbance, d(t), to the controlled variable, while P’s movements are outputs, qo(t), that continuously compensate for those disturbances, keeping the controlled variable in the reference state. The mirror image relationship between disturbance and output is the S-R relationship between these variables

1.6  Confusions and Illusions

13

that is predicted by Eq. (1.3). This relationship is apparently one of the most interesting aspects of the behavior observed in this demo. When people were asked what P was doing in this demo, about 40 percent of them said that P was copying or mirroring E’s movements; that is, they said that P was responding to the stimulus of E’s movements. Another 40 percent said that P was “drawing something,” such as a kangaroo. In this case it was the pattern of P’s compensatory movements – a side effect of compensating for E’s disturbances – that caught people’s eye. Only 1 percent of those asked about the behavior in this demo correctly guessed that P was acting to keep the knot over the dot. The moral of the rubber band demo is that the existence of S-R relationships in the behavior of living systems can “blind” us to the fact that what the system is actually doing is acting to protect a controlled variable from the effects of disturbance. But S-R relationships are also a problem because they can appear to tell us something about how living systems “work” when, in fact, they tell us almost nothing. This can also be seen from Eq. (1.3). The equation shows that the relationship between disturbance and output (S and R) in a living control system depends on the nature of the feedback function (F) – the environmental connection between the system’s output and the controlled variable – and not on the nature of the organism itself. This means that the observed relationship between S (disturbance) and R (output) for a living control system reflects characteristics of the environment in which the system does its controlling, not characteristics of the system itself. When dealing with a living control system, taking observed relationships between S and R to reflect characteristics of the system itself is a mistake that has been dubbed a behavioral illusion (Powers, 1978). 1.6.2  Illusions A behavioral illusion can be demonstrated in a simple control task where the participant is asked to keep an insect (a spider in this case) as far away from themselves as possible.5 The situation is shown in Figure 1.4. The participant sees a spider on the screen which varies in size, giving the impression that it is varying its distance from the participant. The variations in distance are being caused by the spider pushing toward or away from the participant. The participant uses the mouse controller to push back and keep the spider 5

A demonstration of the behavioral illusion can be found at www.mindreadings.com/ControlDemo/ Illusion.html.

14

Living Control Systems

as far away as possible. The spider’s pushes look like a stimulus (S) that is causing the push back response (R) that keeps the spider at a distance. The effect of the push back response on the apparent distance of the spider – the controlled variable, CV, – is shown by the dashed line connecting the mouse to the line that connects the spider to the participant. This is the environmental feedback function, called k in the diagram. This feedback function is shown as the equation below the dashed line in the figure. It says that the effect of the participant’s push back response (R) on the CV is proportional to a constant, k. At the start of the task k is set to 0.5 and in the middle of the task it is changed to 2, making the effect of R on the CV weaker during the first part of the task (k 1 = 0.5 so CV = 0.5*R) than it is in the second part (k2 = 2 so CV = 2*R). The results of the task are shown in the graph on the right in Figure 1.4. These results are precisely what is expected based on Eq. (1.3), which says that the output of a control system (R in this case) will be negatively related to the disturbance to the controlled variable (S in this case), and the function relating S to R will be the inverse of the feedback function (k in this case) relating R to the controlled variable (CV in this case). The predicted relationship between S and R is, therefore, R = −1/k*S. And this is what is seen in Figure 1.4. The graph on the right shows that the participant’s “push back” mouse movements (R) are negatively related to the spider’s pushes (S), which are the disturbance to the CV (the apparent distance of the spider from the participant). And the slope of the relationship between S and R is proportional to the inverse of the feedback function, k. The two lines on the graph show the S-R relationships observed in the two different phases of the experiment with the same participant; in the first phase (Take 1) the feedback function, k 1, was weak (0.5) and in the second (Take 2) the feedback function, k 2, was strong (2). These relationships between the spider’s approach (S) and the participant’s

R = -1/k S

Take 1 High Sensivity (Weak Feedback) Take 2 Low Sensivity (Strong Feedback)

R

S Push Back (R)

R = -1/k2S

CV = k S

R = -1/k1S

a

Spider Push (S)

b

Figure 1.4  An insect avoidance task that illustrates the behavioral illusion.

1.7  What Are You Doing?

15

push back (R) would conventionally be seen as characteristics of the participant, as represented by the equation inside the participant’s head. The steepness of the slopes of these relationships would be taken to indicate the participant’s sensitivity to the spider’s approach; the steeper the slope, the greater the participant’s sensitivity. But by this interpretation the participant’s sensitivity changed dramatically in the middle of the task for no apparent reason; the participant’s sensitivity went from being quite high in the first part of the task (indicated by the steep slope of the light gray line) to being quite low in the second part (indicated by the shallow slope of the black line). It looks like the participant’s fear of spiders declined precipitously in the second half of the experiment for no reason. While that is a possibility, a far more likely explanation is that it was the participant’s environment, and not the participant, that changed in the middle of the task. And, indeed, the change in the slope of the S-R relationship between the first and second parts of the task is exactly what is expected from the change in the environmental feedback function that occurs between the two parts of the task. When the feedback connection between the mouse and the controlled variable was weak, as it was in the first part of the task, the participant had to “push harder” with the mouse to oppose the spider’s pushes than when the feedback connection was strong, as it was in the second part of the task.

1.7  You Can’t Tell What an Organism Is Doing by Looking at What It’s Doing The illusion demonstrated in Figure 1.4 reveals a remarkable fact about the behavior of a living control system. It shows that you can’t understand the behavior of such a system by looking at what it’s doing – in this case, by looking at how the system responds to stimuli. That’s because, per Eq. (1.3), what the system does – how it reacts to stimuli – reflects characteristics of the feedback function connecting the system’s output to the variable it is controlling, not characteristics of the system itself. Even when we find nearly perfect relationships between stimuli and responses, like those in Figure 1.4, those relationships tell you more about the environment in which the system is behaving than about the system itself. This is why the input–output (or S-R) relationships found in conventional psychological studies, including studies based on the MCT mapping of control theory to behavior, tell us little about the properties of the organism itself. While these relationships do tell us about how organisms will react to stimuli in certain situations – something that is good to know when training pilots or

16

Living Control Systems

drivers – they don’t tell us why they react that way. The answer to the “why” question is “in order to protect controlled variables from disturbance.” Both control blindness and the behavioral illusion result from taking the behavior of a living control system at “face value.” It is based on the quite reasonable assumption that you can learn why organisms behave as they do by making careful observations of their overt behavior. This assumption has certainly worked well in the physical sciences. For example, Newton figured out why the planets behave as they do based on the careful observations of planetary behavior made by Tycho Brahe and Johannes Kepler (Koestler, 1990a). Observation of overt behavior works in the physical sciences because the systems studied in those sciences have no purpose. Living organisms, on the other hand, do have purposes and the PCT explanation of how organisms work shows that their overt behavior – in the form of their observable outputs – can provide a very misleading basis for developing an understanding of why they behave as they do. If you can’t understand the behavior of a living control system by looking at how it behaves, how do you understand its behavior? The answer, my friend, is blowing in Eq. (1.4), which says that the behavior of a living control system is organized around the control of perceptual variables: p(t) ≈ r(t). This means that everything a control system does is aimed at bringing perceptual variables to values specified by reference signals inside the system and keeping them there protected from the effects of disturbances. Understanding the behavior of a living control system, then, is largely a matter of figuring out what perceptions it is controlling. Once you know what variables the system is controlling, you also know (per Eq. (1.3)) exactly what it will do in response to any disturbances to those variables. That is, when you know what perceptions the system controls, you know what it will do in order to control them. The problem, as the fellow who wrote “Blowin’ in the Wind” might say, is that these perceptions are only in your head (Dylan, 1963a). Chapter 2 will describe ways to get inside the heads of living control systems to see what perceptions they are controlling.

1.8  Summary Research on purpose is based on the fact that behavior is a control process. This fact is established by the observation that what we call “behaviors” are controlled results of an organism’s actions. Controlled results are consistent results produced in the face of disturbances that should prevent

1.8  Summary

17

such consistency. The ability to produce controlled results is explained by control theory, which shows that the behavior of living organisms is organized around the control of perceptual variables. Thus, the application of control theory to understanding the behavior of organisms has come to be called Perceptual Control Theory or PCT, which shows that in order to understand purposeful behavior it is necessary to determine the perceptual variables that organisms control: controlled variables. Failure to notice the existence of controlled variables can lead an observer to see behavior as a response to stimulus input. What is being seen in this case is the organism’s resistance to disturbances to the controlled variable. PCT shows that observed functional relationships between stimuli (disturbances) and responses (system output) depend mainly on properties of the feedback connection between the organism’s output and the variable(s) it controls and very little on the properties of the organism itself. PCT shows that the main goal of research aimed at understanding the purposeful behavior of organisms – research on purpose – should be the discovery of the perceptual variables that organisms control.

2

Doing Research on Purpose

The perceptual variables a living system controls are the raw material of purpose. For example, Romeo, when controlling a perception of his distance from Juliet, has the purpose of perceiving himself close to Juliet; a person controlling a perception of the relationship between their checkbook and bank statement has the purpose of perceiving a balanced checkbook. So when we do research aimed at determining the perceptual variables an organism is controlling, we are doing research on purpose. And since those of us who are doing this research are living control systems ourselves who have the purpose of understanding the purposes of other living control systems, we are doing research on purpose in both senses of that phrase; we are purposefully doing research whose subject is purpose itself (Marken, 2014). As noted in Chapter 1, the main goal of doing research on purpose is to discover the perceptual variables organisms are controlling when they are carrying out various behaviors. Put simply, we want to know what organisms want to perceive. The problem, of course, is that the perceptions organisms control cannot be experienced by outside observers. One possible way to address this problem is to ask the system under study what perceptions it is controlling. This is essentially what we are doing when we ask “what do you want.” The answer is a description of a desired perception, such as a chocolate cake or a kiss. One problem with this approach, of course, is that it can only work on systems that can talk, which pretty much limits its use to non-infant humans and highly trained apes (Premack and Premack, 1984). Another problem is that the usefulness of this method depends on the truthfulness of the person (or ape) being studied. And even if truthfulness were not an issue, it is not certain that people (or apes) are able to accurately describe the perceptions they actually are controlling. For example, if baseball fielders were asked what perception they are controlling when they catch a fly ball, they are unlikely to say that it’s the vertical optical velocity and horizontal optical displacement of the ball relative to 18

2.1  The Test for Controlled Variables

19

a fixed reference point. But research that will be discussed in Chapter 4 has shown that these are, indeed, the perceptions fielders control when they catch fly balls (Shaffer et al., 2013).

2.1  The Test for Controlled Variables The perceptions people and other organisms control when doing things like catching fly balls were discovered using a nonverbal research methodology based on PCT called the Test for the Controlled Variable or TCV. This approach to determining the perceptions a living control system is controlling is based on the fact that these perceptions will be protected from the effects of disturbances. The researcher who suspects that the system under study is controlling a particular perception can test this suspicion by applying disturbances that would have an effect on that perception if it were not being controlled. If these disturbances actually have little or no effect, it is evident that the perception being disturbed is, indeed, under control. A researcher can’t go into the system and disturb its perceptions directly. In the TCV, what is disturbed is the system’s perception as seen from the point of view of the observer; it’s a perception in the researcher that is thought to correspond to the perception controlled by the system under study. When the researcher is perceiving what the system is perceiving, the researcher’s perception corresponds to the controlled variable, qi(t) in Figure 1.1. This may make it sound like doing the TCV requires some special perceptual capacities. But, in fact, no such capabilities are necessary. Indeed, you have probably seen the TCV being done in old movies where the driver of a car is concerned about the possibility that they are being “tailed.” The way the driver tries to find out whether or not this is the case is by randomly changing direction whenever the opportunity presents itself. If the driver is being tailed then the tailing car will eventually reappear in the rearview mirror after each change of direction. If not, the car that appeared to be doing the tailing will soon disappear. This method of testing to see whether or not you are being tailed turns out to be an example of the TCV. It starts with a guess or hypothesis about the perception being controlled. In this case, the hypothesis is about the perception being controlled by the driver of the car in the rearview mirror. The hypothesis is that that the driver is controlling a perception of following you. If this hypothesis is correct then the following driver will act to protect that perception from disturbances and keep turning up in your rearview mirror; if not, the disturbances will be completely effective

20

Doing Research on Purpose

and the following driver will be seen no more. One obvious way to produce such disturbances is by randomly changing where you are heading. If the car behind you keeps turning up in the rearview mirror after each change (disturbance), it is likely that the driver of that car is, indeed, controlling a perception of following you. In PCT terms, the driver is acting to protect this perception from the direction-change disturbances. If, on the other hand, the tailing car disappears from the rearview mirror, the disturbance was effective in the sense that it had the expected effect if the driver was not controlling a perception of following you. It is still possible that you are not being tailed – that the driver of the car behind you is not controlling a perception of following you – even if the tailing car turns up in the rearview mirror after one or two changes of direction; the driver might be controlling a perception of following a route that just happens to correspond to the route you are taking with your random turns. So an important part of the TCV is to apply a number of disturbances to the hypothesized controlled perception. If there is resistance to every one of these disturbances – if, for example, the following car turns up in the rearview mirror after each random change in direction – you can be confident that you have correctly identified the perception that the driver behind you is controlling; you are, indeed, being tailed. You may find it odd that “following” is referred to as a “perception.” Calling it a “perception” makes it sound like “following” is just a “point of view” or “opinion” of the driver in the car behind you. But the driver behind you is not controlling an opinion but, rather, something that seems quite real – the sight of your car. When testing to see if you are being tailed, “following” is called a “perception” because it refers to a state of affairs in the outside world that is represented inside the behaving system (the driver of the following car, in this case) as a perceptual signal (the variable p(t) in Figure 1.1). “Following” actually refers to a number of different possible states of affairs – perceptions – that the driver might be controlling. For example, “following” could refer to keeping your car constantly in view from behind. Or it could refer to keeping your car only intermittently in view from behind. Or it could refer to keeping your car intermittently in view from any angle. In the above example of using the TCV, we made no effort to determine which of these different perceptions was the one being controlled. All you cared about was whether you were being followed. You were being followed if the driver behind you was controlling any one of these perceptions. But when you use the TCV to do research on purpose, one of your main goals is to find out exactly what perception the system under study is controlling.

2.1  The Test for Controlled Variables

21

2.1.1  The Coin Game A nice way to learn how to get a more exact picture of a perceptual variable the system is controlling is by playing a simple example of the TCV called the “Coin Game” (Powers, 1973b, 2005b, pp. 236–238). All that’s required for this game are two people and four coins of any kind. One person is the participant, P, and the other the experimenter, E. P is asked to arrange the coins so that they exemplify some specific condition or pattern, which is kept secret from E. For example, P might arrange the coins into a rectangular pattern. Or P might arrange the coins in a line so that the dates on the coins increase from left to right. There are clearly many perceptual aspects of the coins that P can choose to control. Once the coins are arranged as P wants, E disturbs them by changing the arrangement of coins in various ways, such as by moving them to a new location or turning them over. After each disturbance, P can correct things by restoring the disturbed coins to their original position. If no correction is needed, P can say “No error.” So P is asked to control a perception, which is some “state of affairs” of the coins; E is to infer the perception P is controlling based on observation of P’s behavior – correcting or not correcting the coins after each disturbance. The Coin Game starts with E developing a hypothesis about the perception P is controlling. This hypothesis must be based on how P initially arranges the coins. If the coins are arranged in a row, E’s first guess might be that S is controlling a perception of the coins being in a straight line. E can test this by moving one or more coins so that they are no longer in a straight line. If P does not act to correct for this disturbance, then P is clearly not trying to keep the coins in a straight line and E must come up with a new hypothesis about the controlled perception. If, however, P does correct for the disturbance, it is evidence that P is controlling a perception of the coins being in a straight line. However, it is not conclusive evidence. P might be controlling a different perception that would require correction after the same disturbance. For example, if you had disturbed the straight line by moving a small coin on the left end above a large one in the middle, that would also have been a disturbance that required correction if P had been controlling for the coins increasing in size from left to right. So E must continue testing, disturbing the “straight line” perception in many different ways, before concluding that P is indeed controlling for a straight line. If any of these disturbances leads to no correction, then the “straight line” hypothesis must be rejected and a new hypothesis formed and tested. The new hypothesis should, of course, be one that is consistent with P’s previous corrections (and noncorrections) of E’s disturbances.

22

Doing Research on Purpose

The Coin Game illustrates the way the TCV is used to get an accurate picture of the perception a system is controlling. It is done by testing hypotheses about the perception being controlled by applying disturbances that should have an effect if the hypothesized perception is not being controlled. When these disturbances are corrected, it is evidence that the hypothesized perception may, indeed, be controlled. The testing ends when E is able to correctly predict which of several disturbances will and will not lead to corrective action. At that point E can be confident that she has identified the perception under control. 2.1.2  Accurate Estimation of the Controlled Variable The same principle of control that was used in testing for the perception being controlled when you suspect that you are being followed is also used to determine what perception is being controlled in the Coin Game. The principle is that a controlled perception is one that is protected from all disturbances that would move it from its reference state. This principle is the basis of the TCV. The only difference between the “car following” and “Coin Game” examples of the TCV is that you get a more accurate picture of the controlled perception with the iterative process used in the Coin Game. The accuracy comes from the fact that hypotheses about the controlled perception are constantly refined when disturbances are found to be effective (not corrected). The person doing the TCV doesn’t necessarily perceive exactly what the system under study perceives. This is because the person doing the TCV is looking at what the system is controlling from what is literally a different perspective; the tester and the person being tested are located in two different bodies and, thus, are experiencing things from different points of view. Nevertheless, the experimenter can get a very accurate picture of the perception being controlled by taking this difference into account. For example, in the rubber band demo shown in Figure 1.3, E can see that P is controlling the distance between the knot and the target dot. But this distance is seen from P’s point of view. And because the knot is slightly above the dot, E will be getting a parallax view of the controlled variable. But by taking this parallax view into account, it was possible to get an exact picture of the perception P was controlling in the demo (Willett et al., 2017). Because there is always a difference in perspective between the tester and the system being tested, the variable the system is controlling has a different name depending on whether it is being described from the perspective of

2.2  The TCV as Mind Reading

23

the tester or the system being tested. From the system’s perspective, this variable is called the controlled perception; from the tester’s perspective, this variable is called the controlled variable. Both terms refer to the same variable but since the testing is done from the tester’s perspective, the method used to determine what variable(s) a system is controlling is called the Test for the Controlled Variable rather than the Test for the Controlled Perception.

2.2  The TCV as Mind Reading Since keeping particular perceptions under control reflects one’s purposes and since those purposes exist only in one’s mind, Powers (1979b) called the TCV “… the nearest approach I know of to mind reading”. In his delightful autobiography the Nobel Prize-winning physicist Richard Feynman describes a magician’s trick that uses a version of the TCV to do what looks very much like mind reading. The trick involved finding some money that a person had hidden prior to the magic show. The mind reader “magically” finds the money by having the person who hid it unknowingly lead the mind reader to it. Feynman describes the method used to do the trick as follows: … you hold onto [the person’s] hand, loosely, and as you move, you jiggle a little bit. You come to an intersection, where you can go forward, to the left, or to the right. You jiggle a little bit to the left, and if it’s incorrect, you feel a certain amount of resistance, because they don’t expect you to move that way. But when you move in the right direction, because they think you might be able to do it, they give way more easily, and there’s no resistance. So you must always be jiggling a little bit, testing out which seems to be the easiest way. (Feynman, 1985, p. 48)

In this version of the TCV, you (the mind reader) are E and the person whose hand you are holding is equivalent to P in the Coin Game. The “jiggles” are disturbances and the “resistance” is P’s effort to correct for those disturbances. E’s hypothesis about the variable P is controlling – the controlled variable – may be tough to see. Feynman describes it as “not expecting the mind reader [E] to move the wrong way.” In TCV terms, we would describe it as “perceiving that E is moving toward the money.” The trick works to the extent that P is, indeed, controlling this variable. If so, P’s resistance to the jiggle disturbances tells E that he is moving in the wrong direction and P’s lack of resistance tells him that he is moving in the right direction. E can read P’s mind and locate the hidden money by following the path of least resistance; the path that achieves P’s purpose of having E move toward the money.

24

Doing Research on Purpose

This mind reading trick relies to a large extent on luck; the luck of picking people who, after hiding the money, will be controlling for a perception of the mind reader being able to find it. But the actual TCV doesn’t depend on luck. As was demonstrated by the Coin Game, if E’s first guess about how to define the controlled variable proves to be wrong (because disturbances to this variable are not resisted), E tries another and then another and another, if necessary, until the correct definition is found. 2.2.1  Keeping Track of a Changing Mind Thus far, the demonstrations of the TCV have been done under the assumption that the organism under study is maintaining a fixed reference for the state of the controlled variable. In the Coin Game, for example, it was assumed that P maintains a fixed reference for the perceptual aspect of the coins that is being controlled. In Feynman’s mind reading trick, it was assumed that the person who hid the money has a fixed reference for a perception of the mind reader finding it. But the use of the TCV is not limited to studying purposeful behavior where the reference for the controlled variable is fixed. The TCV can be used to determine what perceptual variables a system is controlling even if the reference specification for the state of that variable is constantly changing, as it is in many examples of real-life behavior. An online illustration of doing the TCV when the reference for the state of the controlled variable is constantly changing is available in the form of an interactive computer demo called Mind Reading.1 In this demo, you are the system under study (P) and the computer is the researcher (E). The demo starts with a display of three cartoon characters moving in different paths around the screen, as shown in Figure 2.1. You are to pick one of those characters and use the mouse to move it around the screen along an arbitrary path of your choice. That is, you are being asked to move one of the characters “on purpose” and the computer’s job is to determine which it is. The computer’s task is particularly difficult because the mouse moves all three characters at the same time so that all three are moving in arbitrary paths around the screen so that none of the paths looks more “purposeful” than another. It is, therefore, impossible to tell, by simply looking at the behavior of the three characters, which is the one being moved on purpose. In order to tell which character is being moved on purpose, the computer 1

An online version of the Mind Reading demo is available at www.mindreadings.com/ControlDemo/ Mindread.html. The demo is based on work described in Marken (1989).

2.2  The TCV as Mind Reading qo d3 d2

25

qi3

qi2 q o

qo qi1 d1

qo

Figure 2.1  One frame of the animated computer display in the Mind Reading demo.

has to “read your mind” to determine what your purpose is. This is done using the TCV. According to PCT, your purpose in the Mind Reading demo is to control a perceptual variable – the position of one of the three characters – relative to a continuously (and arbitrarily) varying reference specification for the position of that variable. The computer uses the TCV to determine which of these three possible perceptual variables – the positions of the three characters – is the one you are controlling. It does this by looking for a lack of effect of disturbances to each of these three variables. The variable that is least affected by the disturbance would then be identified as the one being controlled. The situation is diagrammed in Figure 2.1. The possible controlled variables are the paths of the three characters, which are indicated by the arrows labelled qi1, qi2, and qi3. The disturbances to each of these variables are indicated by the arrows labeled d1, d2, and d3. And the effect of your output – mouse movement – on each of these variables is qo. Determining which character is being moved on purpose in this version of the TCV is complicated by the fact that the reference specification for the position of the controlled character is continuously changing. This is a problem because “lack of effect of a disturbance” is measured as a change in the state of a controlled variable. If the disturbance doesn’t cause such a change, then the variable is likely under control; if it does then the variable is not under control. But in the Mind Reading demo, a change in the state of the controlled variable could be caused by the disturbance to that variable, by a change in the reference for the state of that variable or both. If what is

26

Doing Research on Purpose

mainly a reference-caused change is taken to be the effect of a disturbance, then a variable that is actually under control could be incorrectly seen as one that is not. The Mind Reading demo solves this problem by looking at the average effect of disturbances over time. Each character in the demo is being “pushed around” by a different time-varying disturbance. The average effect of each disturbance on the position of the character it affects is continuously measured by the correlation between these variables during a sliding ½ second time window. The average correlation during this time window will be close to zero for the actual controlled variable and much greater than zero for the variables that are not controlled. This is because disturbances to the uncontrolled variables are not being resisted. Therefore, most of the variation in these variables is caused by the disturbances, resulting in a relatively high correlation between the uncontrolled variables and the disturbances to them. Disturbances to the controlled variable will be resisted to the extent that those disturbances would cause that variable to move from its varying reference state. The result is that most of the variation in the position of the character that is being controlled is caused by variations in the reference specification while very little of it is caused by the disturbance; the result will be a near zero correlation between the controlled variable and the disturbance to it. The computer “reads your mind” by simultaneously testing three hypotheses about which character you are controlling. The computer tests these hypotheses by continuously computing the correlation between these disturbances and the movements of each character and identifies the controlled character as the one with a correlation that is closest to zero. It may take several seconds for the computer to determine the controlled character. This is the time it takes to continuously compute the correlations involved in the TCV. But once the computer has “read your mind” it will tell you which character you are moving on purpose by changing that character’s identity. At that point, you can start moving a different character on purpose to see if the computer can figure out that you have “changed your mind” (adopted a new purpose). When you do this, the character that you had been moving on purpose should return to its old identity and in a few seconds the computer should be able to determine which character you are now moving on purpose. You can keep changing your mind like this as many times as you wish. How well the computer does at detecting these changes will depend on how well you are able to control the characters in the sense of being able to resist the disturbances to their position and get them to move exactly where you want; the better

2.3  Doing the TCV

27

you get at controlling the characters, the better the computer gets at reading your mind. The Mind Reading demo is meant to be another demonstration of the fact that it is often impossible to tell what a person is “doing” (their purpose) just by looking at what they are “doing” (their overt behavior). That is, it is a demonstration of the fact that you can’t always tell what results a person intends to produce by just looking at the results that happen to catch your eye. If you watch someone doing the Mind Reading demo – someone who is skilled at controlling the position of the characters – you will see that it is impossible to tell with any certainty which of the three characters is being moved intentionally. But the computer, using the TCV, can identify the intentionally moved character with great accuracy. What is being identified is which of three possible perceptual variables – the positions of the three cartoon characters – is being controlled at any one time. But the Mind Reading demo is also a demonstration of a way to use the TCV when the reference for a possible controlled variable is likely to be changing over time. When this is the case, it is still possible to do the TCV as long as you observe the average effect over time of disturbances to the hypothetical controlled variable. You are still looking for lack of effect of the disturbance to the possible controlled variable. But when the reference for the variable may be changing you are looking for a definition of the controlled variable where disturbances produce the least average effect over time.

2.3  Doing the TCV It is possible to describe a general approach to doing the TCV in terms of a formal series of steps, as follows: 1. Develop a hypothesis about the controlled variable by defining a variable that the organism might be controlling. 2. Predict the expected effect of disturbances to the variable, assuming that the variable is not under control. 3. Apply various amounts and directions of these disturbances to the variable. 4. Measure the actual effect of the disturbances. 5. If the actual effect is essentially the same as the predicted effect, stop. The variable is not under control. Return to Step 1. 6. If the actual effect of the disturbances is markedly less than expected, the variable may be under control. 7. Continue applying disturbances to the variable until you are able to reliably predict which disturbances will and will not be opposed.

28

Doing Research on Purpose

8. If all steps of the test are passed, the hypothesized variable is very likely to be a controlled variable and the state in which the variable is being maintained is its reference state. These steps were adapted from the description of the TCV found in Runkel (2003). While they provide a useful guide to doing the TCV, there is more to doing TCV-based research than mechanically following these steps. One must also take into consideration a basic “rule” of all scientific research which is that experiments must be conducted under controlled conditions. In this case, the word “controlled” means “held constant.” When doing the TCV, the disturbances that are used to test hypotheses about a controlled variable should be manipulated while all other variables that could affect that variable are held constant. This should be done for the same reason it is done in conventional psychological experiments: in order to be able to conclude that the results of the experiment are valid. 2.3.1  Eliminate Confounding Variables In conventional experiments, the researcher manipulates an independent variable to see if it has an effect on a dependent variable. A basic requirement for validity is that there are no confounding variables; variables that covary along with the independent variable. When there are confounding variables in an experiment, it is impossible to conclude that the independent variable is the one responsible for any observed effect on the dependent variable. It is in this sense that the results of the experiment are not valid. The same applies to experiments using the TCV. In these experiments, the researcher manipulates a disturbance to see if it has little or no expected effect on a hypothetical controlled variable. The disturbance is equivalent to the independent variable in conventional experiments inasmuch as it is manipulated by the researcher so that the variations in this variable are independent of the actions of the organism under study. As in conventional experiments, confounding variables must be eliminated so that the researcher can have confidence that any observed lack of effect of the disturbance on the hypothetical controlled variable is a result of the disturbance-resisting actions of the organisms and not a result of the effect of a confounding variable. 2.3.2  The Organism Must Be in Control Another important methodological consideration, specific to TCVbased research, is that disturbances be applied in such a way that they

2.3  Doing the TCV

29

can be resisted if the hypothesized controlled variable is actually under control. That is, testing for controlled variables shouldn’t cause the organism to lose control. In the examples of the TCV that have been described so far, this was the case. In the test for whether you were being followed, the driver behind you could easily resist your random direction-change disturbances by making the same changes soon after you do. In the Coin Game, P could easily resist the changes in the coin arrangement by restoring the coins to the intended arrangement. In Feynman’s mind reading trick, the participant could easily resist the mind reader’s gentle “jiggle” disturbances by pushing back against them, telling the mind reader which way to go to find the hidden money. And in the Mind Reading computer demo, the participant can easily resist the disturbances to the position of the character being intentionally moved around the screen. In all these examples of the TCV disturbances are applied in such a way that the person being tested is able to maintain control of the variable being disturbed if, indeed, it is a controlled variable. The reason for this is probably obvious: The TCV can’t reveal the variable being controlled if the organism can’t control it. Determining whether a disturbance will prevent control of a hypothetical controlled variable requires some understanding of what outputs the organism is capable of producing. The researcher should know whether or not the organism is capable of producing outputs that would resist any planned disturbances to the hypothetical controlled variable. In the Mind Reading computer demo, for example, the computer would not be able to tell which character’s movements were being controlled if the disturbances applied to the characters made it impossible for a person to control any of them. In order to use the TCV to successfully determine what variable an organism is controlling the organism being tested must be able to control that variable. 2.3.3  Modeling the Mind The steps of the TCV described earlier represent the basic approach to finding the perceptual variables an organism controls. They describe a process that can achieve the basic aim of the TCV, which is to find variable aspects of the world – perceptual variables – that organisms maintain in reference states, protected from the effects of disturbances. But it is not always possible to carry out the TCV by following the exact steps described earlier, nor is it necessary. Another way to achieve the aims of the TCV is through the use of modeling.

30

Doing Research on Purpose

The modeling approach to the TCV involves building a mathematical or computer model of the behavior under study. This will be a control model, similar to the one shown in Figure 1.1. In this case, the TCV is done by putting different definitions of the controlled variable into the model to see which definition results in model behavior that is most like that of the organism being studied. A simple example of this approach to doing the TCV was described by Marken (2013). The behavior being studied is called pursuit tracking. The situation is illustrated in Figure 2.2(a). A person is asked to keep a cursor, c, aligned with a target, t, that is being moved slowly and irregularly up and down by a ‘driving function’, the disturbance, d. The person keeps the cursor aligned with the moving target by making appropriate movements with the mouse controller. The person who performs this task successfully is controlling a relationship between the target and the cursor, keeping that relationship in the state “aligned,” protected from the disturbance-induced irregular movements of the target. But what is the relationship between the cursor and target that is being controlled? This is the question to be answered by the TCV. There are at least two possible controlled perceptions in this situation, as shown in Figure 2.2(b). The most obvious possibility is the distance between the vertical position of the target and cursor, t – c; the person seems to have the purpose of keeping that perception equal to 0. But another possibility is that the person is controlling the angle from the cursor to the target. This perception can be described as the arcsine of the ratio of vertical to horizontal distance from the target to the cursor: arcsine ([t – c]/s), where s is the horizontal distance between target and cursor. Figure 2.2(b) shows a

b

disturbance, d q.i c

cursor, c

p1 = k (t – c)

I

t-c

t–c

I

t target, t

p2 = arcsin[(t-c)/s]

s t

q.i

t–c

c s

Figure 2.2  (a) A simple pursuit tracking task and (b) two possible controlled perceptions in a pursuit tracking task (adapted from Marken, 2013, Figures 1 and 2)

2.3  Doing the TCV

31

two different input functions that compute different perceptions, which are different functions of the same physical reality – the relative positions of the target and the cursor. The two perceptual variables shown in Figure 2.2(b), p1 and p2, represent two hypotheses regarding the perceptual variable being controlled in the pursuit tracking task. The first step in using modeling to test which of these hypotheses is closest to being correct is to develop a control model of pursuit tracking. A pseudo-code version of a computer program that implements such a model is shown below: 1. 2. 3. 4. 5. 6.

For tick = 1 to Nsamples t : = d[tick] c : = q.o p : = t – c //or arcsine [(t - c)/s] q.o : = q.o + slow*(gain (r - p) - q.o) Next tick

This simple program simulates the behavior of a person doing the pursuit tracking task. The program loops through a tracking trial which occurs over Nsamples of equally spaced time ticks. On each tick of the loop, the target position, t, is set to the current value of the disturbance variable, d[ ], and the cursor position, c, is set to the current value of the output, q.o, which, in this case, corresponds to the position of the mouse. The program then computes the controlled perception, which will be either t – c or arcsine ([t – c]/s. The difference between the current state of the perception and the reference signal, r, is used as the basis for determining the system’s output, q.o. These program statements implement the control system equations presented in Chapter 1. Statement 4 corresponds to Eq. (1.2) – the environment equation – which defines the controlled perceptual variable, p, as a function of disturbances (which are the target movements, t) combined with environmental consequences of the system’s output (which are the cursor movements, c, a consequence of mouse movements). Statement 5 corresponds to Eq. (1.1), the system equation, which defines system output as a function of error – the difference between perceptual and reference signals. The function in this case is a “leaky” integration where the increase or decrease in output, q.o, is proportional (by a slowing factor, slow) to the change in output determined by the size of the error, r − p. The size of the change in output is determined by the gain factor. The output is modeled as a leaky integration in order to take into account the fact that a real control system cannot produce all of the error-determined output instantaneously.

32

Doing Research on Purpose

It also approximates, on a sequential state device (the computer), the fact that all variables in the real system are changing simultaneously. The behavior of this computer model can now be compared to the behavior of a person doing the pursuit tracking task. The aspect of the person’s behavior that is typically compared to the model’s behavior is the person’s output that corresponds to q.o in the model. We are comparing the person’s actual mouse movements to the simulated mouse movements, q.o, made by the computer model. The TCV is done by seeing which version of the model – the one controlling the distance between t and c, t – c, or the one controlling the angle between t and c, arcsine ([t – c]/s) – produces outputs that give the best fit to the person’s behavior. In this case, the distance and angle models fit the data equally well when the horizontal separation between t and c , the value s, was small. However, when s was varied over different experimental runs, the angle model was the clear winner. Indeed, when the fit of the model to human performance is measured in terms of R 2 – the proportion of variance in the person’s output that is accounted for by the output of the model – the R 2 for the model controlling a perception of distance between the target and the cursor was only about 0.5, while the R 2 for the model controlling the angle between the target and the cursor was 0.99.2 Of course, this is not proof that people control the angle rather than distance in a pursuit tracking task. There are still other possibilities, such as a perception of “city block” distance between the cursor and the target (the total of the horizontal and vertical distances between the cursor and target) or a “noisy” version of vertical distance (t – c) that becomes noisier as the horizontal distance increases. Actually, the latter possibility was tested using the pursuit tracking model and produced results that didn’t fit the data as well as the model that controlled the angle from the cursor to the target (Marken, 2013). But this shows that the iterative nature of the TCV described by the sequence of eight steps listed above is also a characteristic of the modeling approach to doing the TCV.

2.4  What Good Is It? At this point, you might be asking yourself “so what?” Beyond knowing that organisms control perceptual variables, what is the TCV good for? What is the significance of finding out which perceptual variables an organism 2

These measures of fit were the result of finding gain and slow parameters that gave the best fit for each version of the model controlling the different perceptual variables.

2.4  What Good Is It?

33

controls? What is the benefit to humanity of knowing this? I think this is a fair question (which was actually asked at this point by one of the reviewers of this text) and I’ll try to answer it the way Michael Faraday is said to have answered a similar question regarding his discovery of electromagnetic induction (Cohen, 1946). After having given a public demonstration of the induction phenomenon, Faraday was asked – possibly by the Prime Minister at the time – “What good is it?” Faraday replied: “What good is a newborn baby?” The point, of course, is that induction, like a newborn, is just a set of possibilities. The newborn may end up being good for curing disease or curing meat, sowing crops, or sewing pants. Electromagnetic induction turned out to be good for building electrical generators and motors, among many other devices essential to the functioning of modern societies. The question “What good is it” could be asked of all basic research in psychology. But it may be hard to imagine what the discoveries of research on purpose could be good for because the aim of this research is to discover what organisms control rather than what controls organisms. When we ask ourselves what a discovery is good for we tend to look for answers that tell us how the discovery will lead to developments that give us greater control over our world. The discovery of electromagnetic induction was certainly good for us in that way; it gave us greater control of the energy that makes societies run. The results of research in psychology are thought to be good for us to the extent that they promise to give us greater control over psychological problems – problems with the way people behave. These kinds of results have come from psychological research that is aimed at finding the causes of behavior. Presumably, when you know what causes some behavior you can control it by manipulating those causes appropriately. Although it might be easier to imagine the beneficial possibilities resulting from research aimed at finding the causes of behavior, there is reason to believe that these discoveries are quite misleading (Powers, 1978). As noted in Chapter 1, if organisms are, indeed, living control systems, then the appearance that their behavior is caused by external events is something of an illusion. It results from ignoring the existence of controlled variables. Thus, the discovery of controlled variables is the sine qua non for understanding the behavior of organisms. Once you have discovered the variables an organism is controlling, you are in a position to develop accurate models of the organism’s behavior. These models could make it possible to solve many of the problems of humanity. For example, they could be used in computer simulations to evaluate the likely effectiveness of policies or treatments aimed at improving human performance when the

34

Doing Research on Purpose

use of randomized controlled trials testing is not feasible. This modeling approach has been used, for example, to evaluate the likely effectiveness of different policies for reducing medical error (Marken, 2005b). A PCTbased control model of behavior has also been used as a basis for designing systems that are compatible with the controlling nature of the human users of these systems (Marken, 1999). And a promising approach to doing psychotherapy, called the Method of Levels or MOL, has been developed in the framework of the PCT model of behavior (Carey, 2006). The usefulness of these approaches to solving problems will depend on the accuracy of the PCT model of behavior; and the accuracy of the model will depend on people doing research on purpose using the TCV to discover the perceptual variables that people (and other organisms) actually control. This is ultimately what doing research on purpose “is good for.”

2.5  Summary Research on purpose is aimed at identifying the perceptual variables that a system is controlling when it is carrying out various purposeful behaviors. The basic methodology used to do this kind of research is the test for controlled variable or TCV. The TCV involves testing hypotheses about the perceptual variables an organism controls. These hypotheses are formal definitions of the perception under control. They are tested by looking for lack of an expected effect of disturbances to the hypothesized controlled variable. The test can be done by converging to a correct definition of the controlled variable through iterative application of disturbances to different hypotheses about the controlled variable or by simultaneously testing these hypotheses in a working model of the behavior under study. But however it is done, the TCV has to start with the formulation of hypotheses about the variables around which behavior is organized. The researcher has to be able to look at what an organism is doing and come up with definitions of the variables it might be controlling – definitions that are precise enough so that it is possible to predict how disturbances would affect these variables if they were not being controlled. Chapter 3 makes some suggestions about how you might go about doing this; that is, it is about how you might go about getting started at doing research on purpose.

3

Getting Started

Research on purpose, like all psychological research, starts with an interest in understanding behavior. It can be a simple behavior like balancing on one foot or a more complex behavior like balancing a budget. But whatever kind of behavior it is that interests you, if it is the behavior of a living organism then it is almost certainly purposeful, which means that it is the behavior of a living control system. The first step in trying to understand the behavior of such systems is to try to come up with ideas – hypotheses – about the perceptual variables the system might be controlling. To do this you have to be able to look at behavior as a control process – through control theory glasses (Marken, 2002a).

3.1  Looking for Controlled Variables The main things we want to see through control theory glasses are the variables the system is maintaining in constant or variable reference states, protected from the effects of disturbances. It can be difficult to see these variables because the disturbances that affect them are often either invisible or too obvious. When disturbances to a controlled variable are invisible, it is hard to see that the system is resisting them; the controlled variable doesn’t look like it is being controlled. An example of this is the behavior “sipping tea,” mentioned in Chapter 1, where the main disturbances, gravity and the forces generated by the movement itself, are invisible, so the person sipping tea seems to be doing nothing more than emitting the sipping movements rather than controlling them. And when disturbances to a controlled variable are too obvious, the system’s resistance to them is what catches the eye; the system appears to be responding to stimuli (the disturbances) rather than acting to keep a variable under control. An example of this was the “control blindness” phenomenon described in Chapter 1 where the participant’s responses to the experimenter’s disturbances blinded observers 35

36

Getting Started

to the fact that the location of the knot in the rubber bands was under control. Control theory glasses are techniques for helping you see controlled variables whether the disturbances to these variables are either invisible or too obvious. 3.1.1  What’s in a Name? One technique that can help you see controlled variables is to look at the name of the behavior under study. These names often provide a hint about the variable that is being controlled. The names of behaviors describe consistently produced results of an organism’s actions. The behavior called “brushing teeth,” for example, is called that because we see people acting to produce the same result – brushed teeth – over and over again. Similarly, the behavior called “making scrambled eggs” is called that because we see people acting to produce the same result – scrambled eggs – consistently. By looking through control theory glasses, we realize that these consistent results are being achieved in the face of disturbances – which are the different circumstances that exist each time these behaviors are carried out; disturbances such as variations in the amount of food in your teeth each time you brush or variations in the heat of the pan each time you scramble the eggs. The consistently produced results for which behaviors are named refer to the reference states of controlled variables, not to the controlled variables themselves. The reference state is the value of a variable; it is not the variable itself. Therefore, in order to use behavior names as hints about controlled variables you have to be able to go from the description of a reference state to a description of the variable of which that reference state is a value. This can often be done by adding modifiers, such as the suffix “ness,” to the description of the reference state in order to turn that description into the name of a variable. “Brushed teeth” can be turned into the reference state of a controlled variable called the “brushed-ness” of the teeth; the behavior “brushing teeth” can then be seen to involve controlling the brushed-ness of your teeth by moving a brush over them until they are in the state “brushed.” Scrambled eggs can be turned into the reference state of a controlled variable called the “scrambled-ness” of the eggs. The behavior “making scrambled eggs” can then be seen to involve controlling the scrambled-ness of eggs by heating and stirring them until they are in the state “scrambled.” In this way, the names of behaviors can be used as lenses to help you see the variables that an organism might be controlling when it is carrying out those behaviors.

3.1  Looking for Controlled Variables

37

3.1.2  Clues from Stimulus–Response Relationships Another place to look for controlled variables is in the existing psychological research literature. But this looking has to be done through control theory glasses. This is because most of the research described in this literature is based on the idea that behavior has causes but no purposes. Therefore, research results are reported in a way that emphasizes the apparent causal nature of the behavior under study, making it difficult to see the controlled variables around which this behavior is likely to be organized. An example of this can be seen in studies of classical conditioning where the behavior being conditioned, such as salivation, is described in causal terms; food powder in the mouth is seen as a stimulus that causes the salivation response (Pavlov, 1928). But through control theory glasses, we can see that we are dealing with the behavior of a living control system. So what appears to be a stimulus is likely to be a disturbance to a controlled variable and what appears to be a response is likely to be the output that compensates for these disturbances. The controlled variable would be something that is being kept in some reference state by the joint effect of the stimulus (disturbance) and response. One possibility in the salivation conditioning studies is the viscosity of the bolus in the mouth that results from the combination of food powder (stimulus) and salvation (response). The dry food powder might be a disturbance to a controlled variable that could be called “swallowability.” Salivation would then be seen as the output of a control system that is aimed at keeping this variable in the state “swallowable.” By looking through control theory glasses, you have identified a possible controlled variable in an experiment where the behavior was seen as a mechanical response to stimuli. But this is only a possible controlled variable. The next step is to use the TCV to test to determine whether the variable “swallowability” is, indeed, being controlled. Whether it is or not, it is probably not the only variable being controlled when salivation is combined with food powder in a dog’s mouth. Some of these possible controlled variables are hinted at by what have been identified as the functions of saliva (Tiwari, 2011). One of these functions is the initiation of the digestion of starch, which suggests the possibility of a chemical-controlled variable, such as the concentration of maltose in the mouth. Maltose is a sugar that results from a chemical reaction between the starches in the “stimulus” food with the enzyme amylase in the salivary “response.” The concentration of maltose is a controlled variable if it is kept at a reference level, protected from disturbances, such as variations in the amount of starch in each bite of the food “stimulus,” by the production of the proper amount of salivation by the salivary glands.

38

Getting Started

Looking through control theory glasses at the stimulus–response – or independent–dependent variable – relationships found in the existing psychological research literature can lead to the identification of possible controlled variables that are much more complex than those involved in the salivary reflex. For example, in his classic studies of “compliance” behavior, Solomon Asch (1951) found that a person can be made to publicly say something is true that he or she privately knows to be false. This happens when the person sees that a majority of others are saying the opposite of what he or she knows to be the case. The person who goes along with others this way is said to be “complying with the majority.” Asch found that the likelihood of compliance increases with the size of the majority. The conclusion from this research was that the size of the majority is a stimulus that causes the compliance response; the greater the stimulus (majority), the greater the response (the chances of compliance). But by looking at these results through control theory glasses, we can see that the size of the majority could be a disturbance to a controlled variable called the “degree to which I look like an oddball” (remember, a controlled variable has to be a variable, hence the “degree”). The people who comply with the majority would then be doing this in order to keep this variable in a reference state that could be called “not looking like an oddball.” The compliance response can then be seen as an output that compensates for the disturbance that is the unanimous majority; you can’t look like an oddball if your response is the same as everyone else’s. If participants in the Asch conformity study were, indeed, controlling something like the “degree to which I look like an oddball,” it would also explain why Asch found so much variability across participants in terms of their likelihood of complying with the majority. Different participants would be likely to have different references for how much they would not want to look like an oddball. Some might have a reference for not looking at all like an oddball. For them, even a small majority saying the opposite of what they knew to be true would require compliance with that majority. Others might have a reference for not looking too much like an oddball, so that it would take a large majority saying the opposite of what they know to be true for them to have to correct for this disturbance by complying. 3.1.3  Doing What You’re Told Another way to look for possible controlled variables in the existing psychological research literature is by looking to see what it was that the participants in the research were instructed to do. In research with humans,

3.1  Looking for Controlled Variables

39

these instructions are typically given verbally and reported in the Method sections of the research report. In studies with nonhuman organisms, these instructions are given implicitly by what are called establishing operations, such as food deprivation and training procedures. These instructions and establishing “operations” tell the participants in behavioral research studies what purpose(s) they are to adopt. In most behavioral research, the participants are instructed to respond in some way to the conditions (stimuli) presented to them by the researcher. In studies of reaction time, for example, participants are instructed to respond as quickly as possible by pressing a button when a stimulus, such as a light, is presented. In survey, research participants are asked to respond by checking one of several possible answers to a stimulus question. In both cases, participants are being asked to have the purpose of responding to certain stimuli. Looking through control theory glasses, we can see that the participants are being asked to control a perception. In the typical reaction time study, for example, the participants are asked to control a perception of the relationship between a light being on or off and a button being pressed or not. This relationship is a perceptual variable that the participant is being asked to keep in a reference state that can be defined as the contingent relationship: if (light = on) then (button = pressed) else (button = not pressed). This contingency is one of the several possible states of this variable, one of the other states being the opposite contingency: if (light = on) then (button = not pressed) else (button = pressed). The light being on or off is the disturbance to this controlled variable. The participant controls this variable, keeping it in the reference state, by pressing the button only when the light comes on. The same kind of analysis can be done with the instructions for completing a survey. The participants in the survey are being asked to control a relationship between the questions and their answers to them. In this case, the relationship to be controlled might be the “correctness” of the answer to the question and the participant is to keep this variable in a reference state that could be described as: (answer|question) = correct, the “|” means “given.” Again, the questions are the disturbances to this controlled variable and the participant controls this variable, keeping it in the reference state by answering the question in a way that is thought to be correct. The variables that participants are instructed to control in conventional psychological experiments are usually not the main attractions of these studies. The main attractions are the relationships between independent and dependent variables. The main attraction in the reaction time study is how

40

Getting Started

long it takes to make the response to the light going on – the response that keeps the controlled variable in the reference state. The main attraction in the survey is to see what kind of answers are made to the survey questions – the answers that keep the controlled variable in the reference state “correct.” However, even though they are not the main attractions, it is a useful exercise to try describing the variables that participants are instructed to control in these experiments. That is, it gives you practice in identifying possible controlled variables and it helps you see the essential role of controlled variables in conventional psychological research. It makes you aware of the fact that conventional research depends on participants being living control systems. If participants did not control the variables they are instructed to control, not much would happen in conventional research studies. If, for example, participants didn’t control for a contingent relationship between light and button press in the reaction time experiment – as they were asked to do – there would be no reaction time to measure since a light doesn’t ordinarily cause people to press a button. And if participants in a survey didn’t control for answering the survey questions correctly, there would be no survey to score, since survey questions don’t ordinarily cause people to answer them. It is also useful to try to describe the variables participants in research are instructed to control because it can make you aware of the different types of perceptual variables around which purposeful behavior is organized. For example, in the reaction time experiment, participants are asked to control a contingency-type perception; in the survey, study participants are asked to control something more like a logical-type perception. These perceptions are different types in the sense that there are many different instances of each type, but all the instances of a type have something in common. For example, in all reaction time experiments, participants are asked to control a contingency-type perception, but the nature of the contingency and the variables involved in it can be quite different. In some reaction time experiments, for example, participants are asked to control a simple contingency like the one described above: if (light = on) then (button = pressed) else (button = not pressed). In others, participants are asked to control more complex contingencies, such as if (light = red) then (press left button) else if (light = green) then (press right button) else (don’t press button). This contingency is more complex, but it is still a contingent relationship between stimulus and responses. The same applies to the logicaltype perception controlled in the survey. In this case, some surveys may ask the participant to pick the only answer that makes the logical variable correct. In others, the participant may be asked to pick all answers that make it correct, so that (answer1 & answer2… & answerN|question) = correct.

3.2  Refining Hypotheses about the Controlled Variable

41

In later chapters, we will see that an important goal of research on purpose is to determine the different types of perceptions that organisms control and whether there is any relationship between these different types of controlled perceptions. This aspect of research on purpose is very similar to the taxonomic classification of organisms in biology and the search for phylogenetic relationships between these classifications. The hope is that this research will provide the basis for a grand organizing principle for classifying controlled variables in the way that evolutionary biology provides the grand organizing principle for classifying organisms.

3.2  Refining Hypotheses about the Controlled Variable Once you are used to looking at behavior in terms of possible controlled variables, you can start trying to develop more precise descriptions of these variables; the kind of descriptions that can be used when doing research on purpose. This kind of research starts with precise descriptions of possible controlled variables. These descriptions are then used as hypotheses in tests to determine what variables an organism is actually controlling. Getting such a description means going from a general description of a possible controlled variable to a more specific one that allows precise measurement of the state of that variable. In conventional psychological research, this is called giving the variable an operational definition – defining the variable in a way in which it can be measured. When doing research on purpose, the best way to approach the development of precise operational definitions of controlled variables is from the point of view of the behaving system, rather than from the point of view of the observer of that system. Since behavior is generally described from the point of view of an observer of the behaving system, the initial description of a possible controlled variable is likely to be from the observer’s perspective. Operationally defining a possible controlled variable from this perspective can result in a misunderstanding of what the system is actually controlling. The problem can be illustrated by looking at a study of the “egg-rolling” behavior of the Graylag goose (Lorenz, 1970). This behavior consists of stereotyped movements that the goose uses to retrieve eggs that have rolled from the nest. The goose reaches out with her neck and hooks the egg with her bill and then rolls the egg back to the nest, balanced behind her bill. When you watch this behavior, it looks like the goose is controlling the position of the egg, moving it carefully back into the nest. But we can get a better idea of what the goose is actually controlling by looking at the situation from the goose’s perspective, through control theory glasses. When we do, we can see that once the goose has the egg behind her

42

Getting Started

bill, all she knows of it is the pressure it exerts against the back of her bill. So the goose can’t be controlling the position of the egg because she can’t see it. She must be controlling something like the pressure exerted by the egg against the back of her bill. Testing this hypothesis would require some pretty sophisticated instrumentation to measure the effect of disturbances to the pressure exerted on the back of the bill by the egg. However, if the goose were controlling this variable, it would explain a phenomenon discovered by the ethologists who studied this behavior. The ethologists found that when the egg is surreptitiously taken from the goose while she is moving it back into the nest, the egg-rolling behavior continues as though the egg was still there. The ethologists saw this as evidence that the egg-rolling behavior is a fixed action pattern: a sequence of neck movement that is caused by the sight of the egg and, once started, runs to completion whether the egg is present or not. If, however, the goose is actually controlling the pressure of the egg against the back of her bill, the neck movements that continue when the egg is gone would be the goose’s futile efforts to restore the desired amount of pressure on the back of her bill after the egg has been removed. A computer simulation shows that this is indeed the way the goose would behave if it were controlling the pressure on the back of its bill.1 This is one piece of evidence that “pressure on the back of the bill” is a more precise a definition of the variable the goose is controlling than is “the position of the egg.”

3.3  Data Collection and Analysis Once you have developed a hypothesis about the perceptual variable or variables an organism might be controlling, you are ready to start testing your hypothesis using the TCV. As in conventional behavioral research, this testing involves collecting and analyzing data. But the way data are collected and analyzed when doing research on purpose differs in important ways from the way it is done in conventional behavioral research. The difference in the way data are collected is in how participants are assigned to the test conditions and in the way data are analyzed to determine the significance of the test results.

1

An online demonstration of the simulation is available at www.mindreadings.com/ControlDemo/ Goose.html.

3.3  Data Collection and Analysis

43

3.3.1  Assigning Participants to Conditions In conventional research, participants are assigned to conditions in groups; either different groups are assigned to each condition (a completely randomized design) or the same group is assigned to all conditions (a repeated measures design). In research on purpose, each individual participant is tested in all conditions. The difference is between testing groups versus individuals. Runkel (1990) calls these two approaches to data collection casting nets and testing specimens, respectively. The casting nets approach to data collection is done under the assumption that behavior is inherently “noisy,” being subject to many causes of random variation. So averaging the behavior of many participants should give a more accurate picture of the processes that produce the observed behavior than does the behavior of any one participant. But it can be shown that this assumption is not necessarily correct; averaging over groups of participants can actually obscure the processes that produce behavior when the behavior is that of living control systems. Powers (1990) demonstrated this by setting up a simulated experiment to test the effect of the size of a reward on the amount of effort expended to get it. The participants in the experiment were a sample of 4,000 computer-simulated control systems. Each control system was controlling for a certain amount of reward by exerting the necessary level of effort. The parameters of each control system – in particular their reference level for the amount of reward and their sensitivity to error (the discrepancy between their reference for reward and the actual amount of reward received) – were randomly selected. The simulation resulted in a measure of the amount of reward received and the level of effort exerted to achieve that reward by each control system on each iteration of the simulation. The results are shown in Figure 3.1. Figure 3.1(a) is a scatter plot of the relationship between the levels of effort expended to get various amounts of reward by each of the 4,000 simulated control systems. The relationship is statistical, but the trend is clear from the regression line; there is a positive relationship between reward and effort, the correlation between these variables being 0.536. While correlation does not imply causality, these results would likely be seen as being produced by organisms that can be made to work harder (exert greater effort) by providing greater rewards. And it would be thought that this characteristic of the organisms would be made clearer if it were possible to reduce the sources of random variation that cause the data to deviate from the regression line.

44

Getting Started

Figure 3.1  Misleading results that could be obtained when using the method of casting nets to study living control systems (adapted from Powers, 1990, Figures 1 and 2).

But we know that these data were not produced by such systems – systems whose behavior is caused by rewards. Rather the data in Figure 3.1 were produced by control systems that were controlling for different reference amounts of reward. Such systems actually put out less effort for increasing amounts of reward. This result is shown in the right two panels of Figure 3.1 for two sets of control systems with different references for the amount of reward desired; one set of control systems with a reference for 293 units of reward and another set with a reference for 107 units of reward. The graphs show that as the amount of reward increases, the amount of effort used to get that reward decreases; for these systems, there is a negative correlation between reward and effort. And these correlations are nearly perfect −0.999 for the systems with a reference of 293 and −0.997 for the systems with a reference of 107 – because this is the way the systems work; the effort put out by a control system is proportional to the amount by which disturbances push the controlled variable – reward in this case – away from its reference value. The apparent positive relationship between reward and effort that is seen in the group data results from the fact that each individual control system has a different reference for the amount of reward it wants. This difference between systems is also responsible for the apparent “noisiness” of the data – the spread of the data points around the regression line – since the range of variation in effort required to bring rewards to their reference is proportional to the size of each system’s reference for those rewards. The results in Figure 3.1 demonstrate the astonishing fact that the method of casting nets – averaging over groups of individuals – can produce a picture of the nature of the systems under study that is precisely the opposite of their actual nature. The actual nature of the system is most clearly revealed

3.3  Data Collection and Analysis

45

using the method of testing specimens – looking at the results from one participant at a time. The group data produce a noisy picture of systems whose behavior is not noisy. The individual data produce a clear picture of systems that are acting to control the amount of reward they received. The results of this simulation show why the preferred approach to doing research on purpose is to test one participant at a time using the method of testing specimens. 3.3.2  Significance: Statistical and Scientific In conventional research using the method of casting nets, results are considered scientifically significant only if they are also statistically significant. These are two very different kinds of significance. Research results are scientifically significant if they tell us something important about the phenomenon under study that we didn’t know before or if they confirm (or reject) the predictions of a theory of the phenomenon under study. Research results are statistically significant if the chances are less than 5 percent or 1 percent that they occurred by chance. Statistical significance is a prerequisite to scientific significance because the results of conventional behavioral research tend to be quite noisy. This noisiness can be seen in the fact that even when an experimental result is found to be statistically significant, the independent variable typically accounts for no more than 34 percent of the variance in the dependent variable (Marken and Horth, 2011). This means that, on average, a statistically significant result is the one where nearly two-thirds of the variance in the behavior under study cannot be explained; it is essentially random noise. When doing research on purpose using the method of testing specimens, results are evaluated mainly in terms of their scientific significance. The results of this research are rarely evaluated in terms of statistical significance because they are not considered to be of scientific interest if the noise level in the data is greater than 1 percent. That is, when we do research on purpose, we expect to find correlations on the order of 0.997. In particular, we expect to find such correlations between the behavior of the organism under study with the behavior of a control system model of the organism. We expect these kinds of correlations because we have consistently been able to find them in studies using the TCV. We get them when we have correctly identified the variables the organism is controlling. If we don’t get correlations on the order of 0.997, we assume that there is something wrong with our hypothesis about the controlled variable, our methods for testing that hypothesis, or our interpretation of the results. We then try adjusting

46

Getting Started

each of these aspects of the research with the aim of getting correlations on the order of 0.997 or higher between model and real behavior. When you find correlations on the order of 0.997 between model and data you have discovered what can be considered a true fact of nature; a fact that can be easily confirmed by anyone who makes the same observations about that behavior. It is not a result that is probably true; it is a result that will always be observed to within a negligible degree of error. There is no concern about whether the result can be replicated (Ioannidis, 2012); it is a result that will always be replicated, like Ohm’s law of the relationship between voltage and current. To date, however, research on purpose has found results of this quality only for very simple behaviors, such as that in the pursuit tracking task described in Figure 2.2. Obtaining high-quality results in experiments involving such simple behaviors may seem trivial compared to the importance of understanding something about more complex behaviors such as prejudice, conformity, and greed – behaviors that affect the quality of one’s own life and that of others. There is no reason that research on purpose can’t be applied to understanding such behaviors. But the bias of those doing research on purpose would be to do studies of these more complex behaviors that produce results that are of the same quality as those obtained in studies of simpler behaviors. This could presumably be done by carefully moving from studies of simple to more complex behavior. Bill Powers put it this way in a post to a Perceptual Control Theory discussion group: If you can get 0.997 in a simple experiment, maybe you can get the same result with a slightly more complicated one. Yes, you can, it turns out. You can even have two people controlling the same display, with two interacting models predicting their behavior, and still get 0.997 correlations. What about four people? What about having them control something a little more complex, with somewhat more complex actions? Yes, and yes. The correlations hold up. The model continues to work… [Research on purpose] is concerned with establishing some facts about behavior that are as accurately known as most facts in physics.… This research is constructing a [scientific foundation for psychology] that is at the same level of detail that Galileo explored by rolling balls down inclined planes and timing pendulums with his pulse…. (Powers, 1992)

Hopefully, this statement will reassure those of you who are interested in starting to do research on purpose that the time spent doing studies of what might seem like trivially simple behaviors is not a waste. Rather, it is time spent developing the skills that can then be applied to doing studies that will reveal true facts about the kinds of behaviors that are of more practical importance.

3.4  Summary

47

3.4  Summary Doing research on purpose starts with learning how to look at purposeful behavior as a process of control. This means learning to see behavior as being organized around the control of perceptual variables. The next step is to try to see what these variables might be. Ideas about what these variables are might be come from the names we commonly use for various behaviors, as well as from the analysis of existing research studies from a control theory perspective. The names given to behaviors typically refer to consistent results of an organism’s actions that are produced in the context of disturbance; that is, they refer to the reference states of a variable that is (or is related to) a controlled variable. Hints about possible controlled variables can also be found in the independent–dependent variable relationships that are found in conventional psychological research. These hints come from recognizing that independent variables are likely to be disturbances to variables that are being kept under control by system outputs that are the dependent variables in these experiments. The instructions given to participants in existing research studies also provide hints about the reference states of variables that participants in the experiment are to control. Once you have an idea of what variable or variables an organism might be controlling, you can refine that idea by looking at the control process from the point of view of the organism itself. The refined hypothesis about the controlled variable is then tested using the method of specimens, which involves testing one organism at a time under all experimental conditions. The results of this research are evaluated in terms of scientific rather than statistical significance. A result is only considered to be of scientific significance if it is virtually 100 percent replicable, which is true when the observed behavior has a correlation of at least 0.997 with the predictions of a model of that behavior.

4

Basic Research on Purpose

Basic research on purpose is aimed at testing the basic PCT model of purposeful behavior shown in Figure 1.1. The model makes some surprising predictions about how purposeful behavior works. The goal of basic research on purpose is to see how well the purposeful behavior of the model corresponds to that of a living organism. The earliest examples of this research were described by Powers in a classic paper on the proper application of control theory to the behavior of living systems (Powers, 1978). This research was performed using a version of a very simple control task called compensatory tracking. Like the pursuit tracking task described earlier, the participant in the compensatory task is asked to keep a cursor aligned with a target by using an output device such as a mouse or joystick. However, in the compensatory task, the target is stationary while the cursor is being pushed back and forth (or up and down in this case) by both a time-varying disturbance and the movement of the participant’s output device. In order to keep the cursor aligned with the target, the participant must move the output device so that it precisely compensates for the effects of the disturbance.

4.1  Relationships between Variables in a Control Loop The typical results of a compensatory tracking task are shown in Figure 4.1. In this particular implementation of the task, the participant moved a handle to compensate for a computer-generated disturbance that pushed the cursor up and down. The graph in Figure 4.1 shows the variations over time in the value of the output (a joystick handle), cursor, and disturbance over time. These variables correspond to the output, qo, input, qi, and disturbance, d (called qd in the graph) variables in Figure 1.1, respectively.1 The results are consistent with the predictions of the PCT model (Eqs. (1.3) and (1.4)). Specifically, per Eq. (1.3), the 1

To simplify, the (t) argument has been left off the variable names.

48

4.1  Relationships between Variables in a Control Loop

49

Figure 4.1  Typical results of a compensatory tracking task for a practiced participant (from Powers, 1978, Figure 4).

participant’s output is precisely opposed to the disturbance: qo ≈ −d.2 And, per Eq. (1.4), the cursor remains almost exactly aligned with the target, where the target or reference position, r, is indicated by the horizontal line: qi ≈ r. The remarkable thing about these results is that the participant’s handle movements precisely oppose the disturbance even though the disturbance itself is invisible. You might think that what is visible is the effect of the disturbance on the cursor; that this effect could be seen in the movement of the cursor that occurs before the participant can move the handle to oppose it. But there is never a time when the movement of the cursor is caused only by the effect of the disturbance. The participant is continuously moving the handle in order to keep the cursor on target so the effect of the disturbance is always mixed with the effect of the participant’s handle movements. That is, qi = qo + d. The fact that the disturbance is not visible in the cursor movements is also demonstrated by the fact that the correlation between disturbance and cursor variations is always close to zero. And the correlation between cursor and output variations is also close to zero. Nevertheless, the participant is making handle movements that are almost exactly opposed to the disturbance; the correlation between the handle movements and the invisible disturbance variations is typically close to −0.99 (Powers, 1979b). 2

This simplified version of Eq. (1.3) results from assuming that the reference signal, r, that specifies the distance between the cursor and the target is zero and from knowing that the feedback function, F, which is the ratio of handle movement to the effect on the cursor, is 1.0.

50

Basic Research on Purpose 4.1.1  Failure of Cause–Effect Model of Control

The lack of correlation between d and qi and between qi and qo makes no sense from the point of view of a cause–effect model of behavior. According to the cause–effect model, the participant’s handle movements are being made in response to the movements of the cursor. So there should be a strong negative correlation between the cursor and handle movements; when the cursor moves up above the target the handle should be moved downwards and vice versa. But the correlation between the cursor and handle movements is always close to zero. This is not the result of looking at the wrong aspects of the cursor movements for the cause of handle movement (Marken, 1980; Marken and Horth, 2011). From the point of view of the cause–effect model, the behavior seen in the compensatory tracking task must be impossible. However, this result is exactly what is predicted by the control model of behavior. In particular, Eq. (1.3) says that when the reference for cursor position is constant, handle movement is predicted to be a function only of the disturbance to the cursor; handle movement is not a function of cursor variations. And Eq. (1.4) says that the position of the cursor is a function only of the reference signal inside the system. So the high negative correlation between handle and disturbance variations is predicted by the PCT model as are the zero correlations between disturbance and cursor variations and between cursor and handle variations. Still it would be nice to know how a control system is able to produce outputs that precisely oppose disturbances that it can’t see. The answer is found in the fact that the outputs of a control system are being made in response to error, qi – r, not in response to input, qi. Assuming that the reference, r, is a constant, error depends only on the value of qi, which is always a simultaneous result of disturbance and output. So at any instant, the size of the error might be mainly a result of the disturbance or of the system’s own output. But whatever the main cause of the error at any instant, the negative feedback loop will keep the control system “doing the right thing” by producing an output that reduces this error. The result will be that the outputs of the system prevent the disturbance from having much of an effect on the controlled variable (cursor position relative to the target, in this case).

4.2  Locus of Control What distinguishes PCT from other applications of control theory to behavior is the assumption that the reference signal, r, that specifies the

4.2  Locus of Control

51

state of the controlled variable is inside the organism. And it is assumed that this reference signal can be autonomously varied by the organism itself. The PCT model shows that the output required to bring the controlled variable to the varying reference states specified by the varying reference signal is mixed with the output required to protect the controlled variable from disturbance. This is shown in Eq. (1.3) which says that qo = r – 1/F*d. Since in a natural environment, disturbance variations are essentially random, variations in an organism’s output will typically provide only a statistical picture of what an organism is doing – a noisy reflection of the input variable it is controlling. This means that in order to tell what an organism is doing you have to look for regularities in its input, not its output. This is illustrated by the results of the compensatory tracking task shown in Figure 4.2. In this task, the participant was asked to keep the cursor moving in some regular pattern; in this case, the participant chose to make the up and down sawtooth pattern shown in the plot at the top of Figure 4.2. In that figure, qi* corresponds to r, the reference specification for the state of

Figure 4.2  Effect of a disturbance on the ability to detect the system’s purpose when the reference for the controlled variable is being autonomously varied (Powers, 1978, Figure 5).

52

Basic Research on Purpose

the controlled variable, qi. The figure shows that before a disturbance is introduced, temporal variations in the handle movement outputs, qo, are virtually the same as the temporal variations in the participant’s reference specification for the sawtooth variations in cursor movements, qi*; the participant’s outputs in this case are a reflection of her purpose. The portion of Figure 4.2 to the right of the vertical dashed line shows what happens to the participant’s outputs when a random disturbance, qd, is added to the cursor. Before the disturbance is introduced, handle movement, qo, is an exact reflection of the reference pattern of cursor movement, qi*. Once the disturbance is introduced, however, it becomes difficult to tell what the participant is doing based on her outputs. But the input, shown as the cursor movement in the plot of qi at the top of Figure 4.2, remains an accurate reflection of the participant’s purpose. Powers (1978) has speculated that the focus on measurement of behavioral output rather than controlled input “… may be why behavioral science so often has to rely on statistical methods to deal with its subject matter” (p. 431). 4.2.1  Deducing the True Value of a Reference Signal Observed variations in the controlled variable presumably reflect variations in the reference signal inside the brain that specifies those variations. But variations in a controlled variable are not an exact reflection of reference signal variations. This is shown in the top trace of Figure 4.2. Even without a disturbance present, as in the trace labeled qi*, the cursor does not trace out the perfect sawtooth pattern that the participant was asked to produce. That is, variations in the controlled variable, even without disturbances affecting them, are not an exact reflection of the participant’s actual purpose – in this case, the sawtooth variations in the reference signal. A basic question for the PCT model is the extent to which variations in a controlled variable are expected to deviate from variations in the controller’s specifications for the state of that variable. In order to answer this question, we have to be able to accurately infer the true value of the controller’s reference specification for the state of the controlled variable. If we were able to do this, we would be able not only to improve the predictive power of the model but also to provide a behavioral basis for neurophysiological research aimed at locating the neural signals that correspond to an organism’s intentions – the reference signals that specify the intended states of controlled variables. Some progress toward getting an accurate measure of the value of a reference signal was made by Powers (1989). His approach to measuring the reference signal is based on modeling. It starts by recognizing that the

4.3  Testing for Controlled Variables

53

reference signal appears in the control model only in the computation of the error signal: e(t) = r(t) – p(t). Thus, the reference signal, r(t), is equal to p(t) + e(t) and its value can be calculated if we can calculate p(t) and e(t). As noted earlier, p(t) can be taken to be equivalent to the controlled variable, qi(t), which, in a tracking task, is the observed distance between the target and the cursor over time. And Powers showed that e(t) is proportional to (qo(t) – qo(t−1))/K, where qo(t) and qo(t−1) are model-produced outputs and K is the gain parameter that produces values of output that give the best fit to the participant’s actual outputs. The value of the reference signal, r(t), can now be calculated as p(t) + e(t). The accuracy of this method of estimating the true value of the reference signal can be checked by running the control model using the estimated value of the reference signal as the values of r(t) in the model. When this is done, the fit of the control model to the behavior of the participant is excellent, suggesting that the estimated value of r(t) gives an accurate picture of the actual value of the reference signal. Powers’s description of an approach to finding the true value of the reference signal was meant to be an example of research aimed at finding the physiological basis of purposeful behavior. The reference signal is presumably the physiological instantiation of the organism’s purpose; the neural signal that specifies the desired state of a perceptual variable. Powers’s PCT-based method of deducing the reference signal provides a behavioral basis for measuring neural signals. Further research by brain scientists is needed to check the physiological accuracy and validity of these measurements.

4.3  Testing for Controlled Variables An important goal of basic research on purpose has been to demonstrate the importance of controlled variables in the behavior of organisms. One such demonstration used the setup shown in Figure 4.3 (Powers, 1978). The participant (the “subject” in the figure) sees four side-by-side cursors, each moving up and down in its own band under the influence of a different random disturbance. A handle moves all four cursors but affects cursors 1 and 3 (C1 and C3) in the opposite direction to C2 and C4. This makes it possible for the participant to control many different aspects of the same display by varying the position of the handle. Participants in this demonstration were first asked to choose one of the four cursors and hold it steady. All participants could do this easily. Participants were then asked to pick some other aspect of the display, not specified by the experimenter, and hold it constant. Powers notes that most participants were initially baffled by

54

Basic Research on Purpose

Figure 4.3  Multiple choices of controlled variables (Powers, 1978, Figure 8).

the request but after they were given some “broad hints,” all were eventually able to see that it was possible to control several aspects of the display other than the position of one of the cursors. For example, the participants saw that it was possible to keep C1 aligned with C2; or to keep C3 aligned with C4; or to keep three of the four cursors lying on a straight line. There were at least sixteen different aspects of the display that could be controlled. The aspect of the display that a participant chose to control – the controlled variable – was determined using a version of the TCV that is based on calculating a measure of control called the stability factor, S, which is calculated as follows:

S = 1 − (Vexp / Vobs )1/ 2 .

(4.1)

In this equation, Vexp is the expected variance and Vobs is the observed variance of the hypothetical controlled variable. The observed variance is easily calculated as the observed variance of the hypothetical controlled variable as defined by the experimenter. For example, if the hypothesis is that the participant is controlling for alignment of C1 and C2 then Vobs is the observed variance of C1–C2 over the period of the experiment. The expected variance is a measure of the variance of the hypothetical controlled variable that would be seen if it were not under control, which is the sum of the variances of all variables that affect the value of the hypothetical controlled variable. The expected variance of C1 – C2 is, therefore, the sum of the variances of the disturbances to C1 and C2 (d1 and d2) as well as the variance of the handle position (h) that combines with these disturbances to affect the positions of C1 and C2: Vexp = Vd1 + Vd2 + Vh.

4.3  Testing for Controlled Variables

55

If the hypothetical controlled variable is, indeed, under control, then Vexp will be much greater than Vobs and S will be a large negative number. S is actually a statistic with a probability distribution similar to the F distribution used in ANOVA. Powers (1978) suggests that “for S several standard deviations more negative than −1, the behaving system can be called an ideal control system.” That is, if the definition of a possible controlled variable results in an S value several standard deviations more negative than −1 then it can be concluded that the behaving system is, indeed, controlling that variable. In the context of the “multiple-choice” experiment described in Figure 4.3, the stability factor can be used to see which of the different aspects of the display the participant is controlling. This is done by computing S for each of the different definitions of the possible controlled variable. When the correct definition of the controlled variable is found, the result will be an S value that is several standard deviations more negative than that for all the others; this is the definition of the controlled variable that passes the test for the controlled variable. 4.3.1  Two-Dimensional Control Another demonstration of testing for the controlled variable was done in a two-dimensional version of the compensatory tracking task (Marken, 1991). In this task, the participant is asked to keep a cursor aligned with a target in the center of the display screen while the position of the cursor is being affected simultaneously by two different random disturbances; one disturbance pushes the cursor right and left in the horizontal (x) dimension of the screen and the other disturbance pushes the cursor up and down in the vertical (y) dimension of the screen. The participant can compensate for the disturbances and keep the cursor aligned with the target by moving a mouse controller. Horizontal (left–right) mouse movements compensate for the disturbance to the cursor in the x dimension and vertical (forward– backward) mouse movements compensate for the disturbance to the cursor in the y dimension. This experiment was set up to test various theories of how people produce coordinated movements, in this case coordinating the left–right and forward–back movement of the mouse. But it also provided an opportunity to determine what perceptual variables the participant was controlling. The first hypothesis was that the participant was controlling a perception of the cursor in x −y or Cartesian coordinates. This hypothesis was tested with a computer model of two control systems, one controlling the cursor in the x dimension and the other controlling the cursor in the y dimension. This

56

Basic Research on Purpose

model produced data that matched the participant’s data almost exactly; the correlation between the mouse movements made by the human participant and the computer model controlling under the same circumstances (same random disturbances) was nearly perfect (0.997). This looked like good evidence that the participant was controlling a perception of the location of the cursor in x − y or Cartesian coordinates. But there was still at least one other possibility: the participant could be controlling in polar rather than Cartesian coordinates. In polar coordinates, the position of the cursor would be represented by its angle and distance relative to the target. A polar version of the control model of this tracking task fits the data just as well as the Cartesian version. Clearly, both hypotheses about the perceptual variables being controlled in this task can’t be correct. The way to find out which hypothesis is correct was to introduce a disturbance that influenced one perceptual variable but not the other. This was done by applying a disturbance to the cursor in just the x dimension. If the participant is controlling in Cartesian coordinates, a disturbance in one dimension will lead to compensating mouse movements in only that dimension. However, if the participant is controlling in polar coordinates, a disturbance to this one dimension will lead to compensating mouse movements in two dimensions. This is because the polar coordinates of the cursor are both a function of x. It was found that the participant, like the model controlling in Cartesian coordinates, compensated for the one-dimensional disturbance in only one dimension; the behavior of the Cartesian model again matched that of the human nearly perfectly; the polar model did not. This simple two-dimensional tracking experiment shows how accurately a PCT model can predict behavior (r = 0.997) when the perceptual variables the model controls are the same as those controlled by the real system (in this case, a human participant). It also shows how modeling can be used to get the best description of the variables being controlled by the real system. And the behavior in this experiment demonstrates one of the most important characteristics of the PCT model of purposeful behavior: the behavior of the model – it’s observable output – depends on what perceptual variable or variables it is controlling. This fact about the way purposeful behavior works is the reason why research based on PCT is focused on discovering the perceptual variables organisms control rather than the causes of the motor outputs they produce. It is interesting to note that the results of this two-dimensional tracking task seem to be inconsistent with those found in the one-dimensional tracking task described in Chapter 2 (Figure 2.2), where it was found that angle (one

4.4  Beyond Compensatory Tracking

57

of the variables controlled by the polar model) rather than distance (one of the variables controlled by the Cartesian model) was a better description of the variable controlled in that task. The finding of different controlled variables in the two tasks may have resulted from two important differences between the tasks. One is that in the one-dimensional task, the participant was to vertically align the cursor with a target that was a fixed horizontal distance from it, whereas in the two-dimensional task, the participant was to keep the cursor aligned with the target both vertically and horizontally. The other is that in the one-dimensional task only one degree of freedom of output was used to control the one dimension of cursor variation, whereas in the two-dimensional task two degrees of freedom of output were used to control the two dimensions of cursor variation. Either or both of these differences could account for the fact the difference in the variables controlled in the two tasks. Resolving this apparent discrepancy regarding the variables found to be controlled in a one-dimensional and two-dimensional tracking task would be a good topic for a future research project.

4.4  Beyond Compensatory Tracking The advantage of testing the PCT model of purposeful behavior using a tracking task is that the relationship between the observed behavioral variables (qi, qo, and d) and the variables in the PCT model of the behavior in this task (p and r) is clear; qi is the position of the cursor, qo is the movement of the handle or mouse controller, and d is the computergenerated noise waveform that is added to qi; p is proportional to qi and qo is a function of r– p. The disadvantage of testing the model this way is that it seems very artificial. And it is, in the sense that most of us probably don’t spend a lot of time doing compensatory tracking tasks in our daily life. But then balls rolling down inclined planes were a pretty artificial example of motion that, nevertheless, provided a basis for Newton’s laws of all motion, natural (like that of the planets) and artificial (like that of a clock pendulum). Nevertheless, while the compensatory tracking task provides a clear demonstration of how the variables in a control loop behave, its artificiality can make it difficult to see where these variables fit into behaviors that don’t look anything like compensatory tracking. 4.4.1  Operant Analog of Compensatory Tracking One example of basic research on purpose using a task that, at first glance, might not seem obviously equivalent to the compensatory tracking task is

58

Basic Research on Purpose

the operant conditioning experiment (Pierce and Cheney, 2013). One example of an operant conditioning experiment is the shock avoidance task where the participant, typically a white rat, must respond in a particular way to avoid getting shocked. Such an experiment was conducted by Verhave (1959) and subsequently analyzed from a PCT perspective by Powers (1971). In the experiment, rats had to press a lever a fixed number of times in a specified interval in order to avoid a shock. After sufficient training at a given interval setting, each rat approached an equilibrium rate of lever pressing that allowed them to almost completely avoid getting shocked. The experiment was not designed to test the PCT model of behavior. However, the rats’ behavior in the experiment is clearly purposeful; the rats were controlling something. This “something” is equivalent to the controlled variable, qi, in the compensatory tracking task. A reasonable first guess is that the controlled variable in this experiment is the probability of getting shocked, a variable that ranges from 0 (never) to 1.0 (always). The output that affects this variable – the equivalent of qo, in the compensatory tracking task – is the rate of lever pressing; the faster the rate of pressing, the more likely that the required number of presses will be made in the set interval and the less likely it is that a shock will occur. The interval setting is the equivalent of the disturbance, d, in the compensatory tracking task since the probability of getting shocked is a function of the combined effect of press rate, qo, and interval setting, d; the shorter the interval setting, the higher the press rate must be in order for the required number of presses to occur within the interval. In order to do a PCT analysis of the behavior in this experiment, Powers first had to come up with a hypothesis about the variable the rats were controlling. Powers’s initial hypothesis was that the rats were controlling the probability of getting shocked. The next step was to determine how the probability of getting shocked was affected by the rats’ output (lever press rate) and the disturbance (the duration of the interval during which the rat had to make the required number of presses). That is, Powers had to derive an equation that shows how the state of the hypothetical controlled variable, qi, depends on the rat’s output, qo, and environmental disturbances, d. The result was an environment equation (Eq. (1.2) in Chapter 1) of the form qi = g(qo,d). The function g() turned out to be statistical because the interval between lever presses varies somewhat randomly over time; even when the average rate of lever pressing is sufficient to prevent a shock, there will be occasions when the actual rate of pressing is too low and a shock will occur. The next step was to develop the model of the rat itself. This involved developing a system equation (Eq. (1.1) in Chapter 1) that describes how

4.4  Beyond Compensatory Tracking

59

the lever press rate varies as a function of error – the difference between the perceived and desired state of the controlled variable. The perceived state of the controlled variable was a perceptual variable, p, proportional to the actual probability of a shock. The desired state of the controlled variable was a constant reference specification, r, equal to zero. So the system equation was qo = K*(p − r) or simply qo = K*p; the lever press rate is proportional to the perceived probability of getting shocked. The system and environment equations together define the PCT model of the rat’s behavior in the shock avoidance experiment. The simultaneous solution of these equations yields a prediction of how the rat’s lever press rate (output, qo) will vary as a function of the different settings of the interval timer (disturbance, d) under the assumption that the rat is controlling a particular variable. This control model was tested by seeing how well it predicts the observed lever press rate, given the different settings of the interval timer. The results are shown in Figure 4.4. Figure 4.4 shows the lever press rate as a function of time interval for two different rats. Figure 4.4(a) shows the results for a rat that was required to make eight presses during an interval in order to avoid shock and Figure 4.4(b) shows the results for a rat that was required to make only one press during an interval in order to avoid a shock. In both graphs, the actual relationship between press rate and interval duration (shown by the black circles) is compared to that produced by different versions of the PCT model (shown by the open triangles and squares). The open triangles labeled “Model with qi = shock prob” are the average press rates at each timer interval that are predicted by the PCT model under Powers’s initial assumption that the perceptual variable controlled by the rat is shock probability. The open squares labeled “Model with qi = shock rate” are the 8 Press Requirement

1 Press Requirement 15.00

90.00

12.00 Actual

60.00 45.00

Model with qi = shock prob

30.00 15.00

Model with qi = shock rate

0.00 1

2

3

4

5

Interval (min)

a

6

Press Rate

Press Rate

75.00

9.00

Actual

6.00 Model with qi = shock prob

3.00 0.00 1

2

3

4

Interval (min)

b

Figure 4.4  Predictions of the PCT model of the controlling done by rats in a shock avoidance operant conditioning experiment (adapted from Powers, 1971, Table 1).

60

Basic Research on Purpose

average press rates at each timer interval that are predicted by the PCT model under the assumption that the perceptual variable controlled by the rat is shock rate rather than shock probability. 4.4.2  Goodness of Fit: r versus rms Powers ended up testing two models of shock avoidance behavior for both rats but only the results for the rat required to make eight presses in the interval were reported. The models differed only in their assumptions about the perceptual variable controlled by the rats. The fit of the predictions of both models to the data was measured in terms of correlation (r) and root mean square (rms) deviation. The measures of fit in terms of correlation showed that both models fit the data extremely well. The correlations between model predictions and actual behavior for the rats in both press requirement conditions were greater than 0.99; the correlation between the model and actual press rate when qi = shock probability was 0.999 while the correlation when qi = shock rate was 0.998. These correlations imply that the different hypotheses about the controlled variable make no difference in the fit of the model to the data. But, while it is somewhat difficult to see in Figure 4.4(a), the model with qi = shock probability produces press rates that are somewhat closer to the actual press rates than are the press rates produced by the model with qi = shock rate. This demonstrates one of the shortcomings of using correlation as a measure of goodness of fit. A model can produce predictions that correlate extremely well with the data but are actually offset from the data by some constant amount. The fact that this is happening here can be seen by comparing the rms measure of fit for the two controlled variables. The rms value is a measure of the average squared deviation of model behavior from observed behavior in units of the data being predicted (press/min in this case). The rms deviation of model behavior from the data when the rat was assumed to be controlling shock probability was 1.17 presses/min; the rms deviation of model from data when the rat was assumed to be controlling shock rate was 3.21 presses/min. The difference in fit between the model controlling shock probability and the model controlling shock rate is 2.04 presses/min, which is large enough (6 percent of the average response presses/min observed in the experiment) to warrant preferring shock probability to shock rate as the definition of the perceptual variable being controlled by the rats. The clear superiority of the shock probability model did not show up in the correlations because the models differed only in how far their behavior was offset from, not in how well it correlates

4.5  To Study Control the Behaving System Must Be in Control

61

with, the actual data. The fact that shock probability is better than shock rate as a definition of the variable the rats are controlling suggests that rats don’t care about the rate at which they get shocked; they only care about whether they get shocked.

4.5  To Study Control the Behaving System Must Be in Control The shock avoidance experiment is an example of an operant conditioning experiment where the goal is to study the effects of schedules of reinforcement on behavior (Ferster and Skinner, 1957). These experiments were developed in the context of a causal model of behavior; it was assumed that consequences (reinforcements) controlled (or selected) behavior rather than vice versa (Marken, 1985). Schedules of reinforcement were thought to affect the way consequences exert control over the organism’s behavior. The possibility that the opposite might be going on – that schedules might affect the way organisms can exert control over the consequences of their behavior – was rarely considered. Therefore, studies of schedules of reinforcement were often set up in a way that made it difficult for organisms to control the consequences of their actions – what we call controlled variables. When this is the case, it is impossible to test a control model of the behavior because the organism is not in control. The problem is illustrated by the behavior of organisms in a schedule of reinforcement experiment using what is called a ratio schedule. In these experiments, an organism has to make a certain number of responses in order to get reinforcements (typically food pellets). The ratio schedule determines how many responses must be made in order to get the reinforcements. From a control theory perspective, the organism in a ratio schedule experiment is controlling (or trying to control) a variable aspect of the reinforcements, such as their probability or rate of arrival. The different ratios are disturbances that must be “resisted” by responding as many times as necessary to produce each reinforcement. If things are made too difficult, such as by requiring too many responses per reinforcement, then it will look like the organism is not controlling anything and that a control model of the organism’s behavior is not applicable. However, when organisms are tested in a ratio schedule experiment where they are able to control the reinforcements you get results like those in Figure 4.5 (Yin, 2013). The results in Figure 4.5 show the behavior of rats in a version of a ratio schedule experiment called random ratio (RR) where the ratio of responses required per reinforcement varies randomly around an average value, indicated by the number after RR. The conditions of the

62

Basic Research on Purpose 128

Number per min

64 32 16 8

Press Rate Reward Rate

4 2 1

Figure 4.5  Control of rate of reinforcement (reward) in a random ratio (RR) schedule experiment.

experiment required the rats to make an average of five (RR5), ten (RR10), or twenty (RR20) responses to get a reinforcement. Most importantly, the experiment was set up so that the rats were capable of controlling the rate at which reinforcements were delivered. The researchers had hypothesized that rats in ratio schedule experiments are controlling the reinforcement rate. And the results show that when rats are able to control the reinforcement rate, they do. This is indicated by the fact that the reinforcement rate stays nearly constant (at about 3.8 reinforcement/min) despite variations in the disturbance, which are the variations in the average number of responses required to get a reinforcement under the different ratio schedules; the rats are varying their output (response rate) to precisely compensate for these disturbances and keep the reinforcement rate in a constant reference state. The results of this experiment don’t prove that the rats were controlling the rate of reinforcement. The reinforcements were food pellets of a particular size so the rats might have been controlling a variable related to the rate at which the food pellets were delivered, such as the rate at which a particular amount of food is delivered in grams per minute. This hypothesis would have to be tested in another experiment where the size of the food pellets is varied as a possible disturbance to the controlled variable. But the results of this experiment do show that basic research on purpose – research aimed at discovering the perceptual variables that organisms control – can be done using methods, such as operant conditioning, that were not designed to be used for that purpose. They also show that research on purpose must be

4.6  Getting Real

63

done in a way that allows organisms to achieve their purposes. You can’t tell what perceptions organisms control if they are unable to control them.

4.6  Getting Real The operant conditioning experiments described earlier show how the PCT model is mapped to behavior other than that in the compensatory tracking task. But the behavior in these experiments is still somewhat artificial since it occurs in a laboratory situation. So we will now take a look at an example of basic research on controlling that occurs outside the lab to see how PCT maps to behavior that occurs in the “real world.” In this case, the “real world” is a baseball stadium and the research we will look at is aimed at understanding how baseball outfielders catch fly balls (e.g., McBeath, Shaffer, and Kaiser, 1995). Catching a fly ball is clearly a purposeful behavior; the goal is to catch the ball and the fielder achieves this goal in the face of disturbances, which are the variations in the ball’s trajectory as it moves through the air. This behavior is clearly an example of control but it might not be obvious how the PCT model of control maps to it. In particular, it might be difficult to tell what variable the fielder might be controlling. Hints about what this variable might be come from noticing that the fielder typically “keeps an eye on the ball” while moving to make the catch (the famous Willie Mays over the head catch with his back to the batter being a marvelous exception). So the fielder is likely to be controlling a visual variable. One of the first guesses about what that variable might be was made by a physicist who noticed that a fielder could get to the right place to catch a ball by moving so as to prevent the visual image of the ball from accelerating upwards or downwards (Chapman, 1968). In other words, Chapman was suggesting that catching a fly ball involves controlling the vertical optical acceleration of the image of the ball, keeping this variable equal to zero. Chapman’s suggestion, called the optical acceleration cancellation (OAC) hypothesis, provides a nice starting point for building a PCT model of catching a fly ball. It proposes that vertical optical acceleration is the variable fielders control when attempting to catch a fly ball. The rest of the PCT model can now be easily mapped to fly ball catching behavior, as shown in Figure 4.6. The model is organized around the controlled variable, qi, which is defined as vertical optical acceleration by the perceptual function, pv = d2α/dt, where α is the vertical optical angle of the ball and d2α/dt is the acceleration of this angle. The outputs that affect the controlled variable, called ox, are the fielder’s movements toward or away from the

64

Basic Research on Purpose

ball. The disturbances that affect the controlled variable, called dx, are things like the wind and the force of gravity that affect the trajectory of the ball. This completes the environmental side of the PCT model. The model of the fielder is made up of the brain functions that implement the control system that keeps vertical optical acceleration of the ball in a reference state of zero – accelerating neither upward nor downward. The perceptual function produces a perceptual signal, pv, that is an analog of vertical optical acceleration; the comparator computes the difference between pv and the reference signal, rv. The computed difference between pv and rv is the error signal, ev, which drives the fielder’s output, ox, via the output function ox = kev. The control model in Figure 4.6 is drawn in a way that is equivalent to the basic control model shown in Figure 1.1 in order to show how the model that maps to behavior in simple laboratory tasks like compensatory tracking also maps to the behavior in more realistic tasks like catching fly balls. But it is an incomplete model of catching fly balls because it will only catch balls that are hit straight at the fielder. If the ball is hit off to the right or left of the model fielder, it won’t be caught. What is missing from the model is the ability to move laterally – to the left or right – as necessary in order to get to a ball that is hit to one side or the other. This capability

Figure 4.6  Perceptual Control Theory (PCT) model of catching fly balls hit directly at the fielder.

4.6  Getting Real

65

Figure 4.7  Perceptual Control Theory (PCT) model of catching fly balls hit in any direction relative to the fielder (adapted from Shaffer et al., 2013, Figure 5).

can be added by adding a second control system to the model. This control system, which is shown in Figure 4.7, controls the lateral optical angle of the ball with respect to the fielder; this is the deviation of the image of the ball from “straight ahead.” This variable, called β, is controlled relative to a reference of zero (rl = 0), which means that the control system tries to keep the ball directly in front of itself. It does this by producing outputs, oy, that move the fielder to the left or right. Like the first control system that controls vertical optical acceleration, this second control system is functionally identical to the basic PCT control model shown in Figure 1.1. These two control systems operate independently, but they automatically coordinate their activities so that what each system does to control its own perception doesn’t interfere with what the other system does to control its own perception. This coordination is rather amazing since the outputs of each system do influence the perception controlled by the other; the forward and back movements of the system controlling vertical optical acceleration affect the value of the lateral optical angle controlled by the other system, and vice versa. There is no interference because the effect of each system’s output on the perceptual variable controlled by the other is treated as a disturbance and is resisted in the way all disturbances are automatically resisted by the outputs of each control system.

66

Basic Research on Purpose

The model in Figure 4.7 can now account for the behavior of a real fielder catching real fly balls that are hit in any direction relative to the fielder (Marken, 2001). It can also account for the fact that real fielders (and dogs) can catch things that move in irregular trajectories (like fly balls that are hit in a windy stadium or Frisbees thrown to dog in any conditions) (Shaffer et al., 2004; Marken, 2005a). Indeed, the model in Figure 4.7 has been used to account for behaviors that involve catching (or intercepting) objects that move in many different trajectories, such as toy helicopters (Shaffer et al., 2013) and footballs thrown to oneself (Shaffer et al., 2015). In order to get the model to account for the behaviors in these different situations, it was necessary to get the model to control the right variables; the ones that gave the best fit of the model to the behavior. This was achieved by running the model with different mathematical definitions of the controlled variables to see which resulted in the best fit to all behaviors. The result was that a model controlling vertical optical velocity, dα/dt (rather than vertical optical acceleration, d2α/dt), and lateral optical angle (β) gave the best fit to all examples of object interception behavior. These are the controlled variables defined by the perceptual functions in Figure 4.7. The success of this model of object interception shows that studies of realistic examples of purposeful behavior can be used to determine that variables people (and canines) control when they are engaged in these behaviors.

4.7  Summary Basic research on purpose is aimed at testing the PCT model of purposeful behavior. These tests are performed to validate detailed predictions about the relationships between the variables involved in purposeful behavior. They have confirmed some surprising predictions of the PCT model about the relationship between input and output variables that are inconsistent with a cause–effect model of behavior. The main focus of basic research on purpose is testing for controlled variables using some version of the TCV. Laboratory tests for the controlled variable are typically conducted using versions of the compensatory tracking task, where all of the variables involved in the behavior can be readily identified and monitored. Other approaches to testing for controlled variables in the laboratory have used operant conditioning tasks which have been set up so that the system under study is capable of controlling the consequences of its actions. It is also possible to do basic research on purpose in more naturalistic settings outside the laboratory, which allows the study of more “realistic” behaviors – behaviors that are more commonly seen in everyday life, such as catching

4.7  Summary

67

fly balls and Frisbees. In this case, the behaviors under study will typically involve the control of more than one perpetual variable. It is still possible to test for controlled variables in this situation using models that include more than one control system, each controlling a different perceptual variable.

5

Exploring the Hierarchy

Organisms control many different perceptual variables, some at the same time and some one after another. For example, the fielder controls two optical variables at the same time to get to where the ball will land but then must control other variables, such as proprioceptive variables in the arm and hand, in order to get the ball into the glove. So another important question for research on Perceptual Control Theory (PCT) is how all these control systems are organized relative to each other. PCT assumes that the control systems responsible for the purposeful behavior of organisms are arranged in a hierarchy, where higher level systems use lower level systems to achieve their goals (Powers, Clark, and McFarland, 1960a, b; Powers, 1973b). The hypothesis is based on the observation that some purposes appear to be carried out as the means of achieving others. So we carry out the purpose of pressing a button in order to achieve the purpose of turning on the computer; we carry out the purpose of turning on the computer in order to achieve the purpose of checking our email; we carry out the purpose of checking our email in order to achieve the purpose of seeing if there was any reply to our earlier email, and so on. Achieving each of these purposes involves controlling for a specific state of a perceptual variable (or variables); the button press involves controlling for a certain amount of pressure on the fingertip, turning on the computer involves controlling for a particular display appearing on the screen and checking email involves controlling for a display of the contents of the inbox. This implies that higher level control systems control their perceptions by telling lower level systems what perceptions to control. PCT assumes that higher level systems do this by setting the reference inputs to lower level systems. The result is the hierarchical arrangement of control systems shown in Figure 5.1.

5.1  Hierarchical Control of Perception Figure 5.1 is a schematic representation of a hierarchical arrangement of control systems. There are four levels of control systems with six control 68

5.1  Hierarchical Control of Perception

69

Figure 5.1  Hierarchical control model of behavior (adapted from Runkel, 2003, Figure 18-3).

systems at each level. Each control system is made up of three boxes representing the three functional components of the system: the perceptual or input function, I; the comparator, C, and the output function, O. The comparator in each system has a reference signal input, r, that is

70

Exploring the Hierarchy

proportional to the combined outputs of several higher level systems. This can be seen most clearly in the bottom part of the figure which shows only the outputs of the level 2 systems that become the reference inputs to the level 1 systems. The outputs of systems at all levels above level 1 become the references for lower level systems. The outputs of the level one systems go directly into the environment so each of these systems is analogous to the basic PCT control system shown in Figure 1.1. These level 1 systems control their perceptions by producing outputs that have a direct effect on the environment. Systems above level 1 control their perceptions by producing outputs that set the references for the perceptual variables controlled by lower level control systems. In other words, systems above level 1 control their perceptions by telling lower level systems what to perceive, not what to do. Systems at level 1 control perceptions that are a direct function of variables in the environment. Higher level systems control perceptions that are functions of the perceptions controlled by lower level systems. The relationship between the perceptions controlled at different levels of the hierarchy can be seen most clearly in the relationship between systems at levels 2 and 3. Only connections between perceptions (inputs) at these levels are shown. What can be seen is that the perceptual variables controlled by systems at level 3 are a function of several perceptual variables controlled by systems at level 2. Therefore, perceptual variables controlled by higher level systems are more complex (and more abstract) than those at lower levels. Indeed, an important assumption of the model is that systems at each level of the hierarchy control a different type of perceptual variable, ranging from relatively simple types of variables at the lowest levels of the hierarchy to very complex types at the highest levels. The difference in the complexity of the types of perceptual variables controlled at different levels of the hierarchy is meant to account for the difference in the complexity of the purposeful behaviors that organisms – and particularly human organisms – produce. The pupillary reflex, for example, involves control of a much simpler perceptual variable (a perception of the intensity of light) than that controlled when playing chess (a perception of a winning principle or strategy). The hierarchical model in Figure 5.1 looks good on paper but does it actually work? That is, can all the control systems in this hierarchical structure successfully control their perceptions, and do this all at the same time without interfering with each other. And the answer is “yes, they can.” This has been demonstrated using computer simulation (Powers,

5.2  Hierarchical Nesting of Control Systems

71

1979a; Marken, 1990).1 It has also been demonstrated by the development of hierarchical perceptual control systems that can balance an inverted pendulum (Powers, 2005a; pp. 164–166) and produce coordinated movement (Marken, 1986). In all of these models, the perceptions controlled at each level were of different types. These demonstrations show that a hierarchical organization of control systems like that shown in Figure 5.1 – one where higher level systems control perceptions by telling lower level systems what to perceive – is at least a feasible model of the processes that produce purposeful behavior. Whether it is the correct model, however, is a question to be answered by the results of research.

5.2  Hierarchical Nesting of Control Systems 5.2.1  Portable Demonstrations of Hierarchical Control The first tests of hierarchical control were described by Powers, Clark, and McFarland (1960b). They called these tests “portable demonstrations” because they required no special apparatus. The following is their description of a demonstration of a lower level, position control system, S1, operating within the reaction time (“latent period”) of a higher level, movement control system, S2. S [the participant] extends his arm in front of himself, with instructions to hold it steady, and E places his hand lightly on top of S’s. E [the experimenter] gives a sudden sharp downward push, and S’s arm appears to rebound as if on a spring. An electromyograph verifies that this is an active innervated correction and not simply muscle elasticity… E now instructs S to extend his hand as before, and E places his hand on top of S’s. Now E tells S to swing his arm downward as rapidly as he can, as soon as he feels E push down. E’s hand must begin in contact with S’s to make the push as sharp and unexpected as possible. Immediately after the push, S1 [level 1 systems] returns the arm to its initial position, because they act within the latent period of S2 [level 2 systems]. Then, after the return swing is nearly completed, the S2 react by resetting the [reference signals] for the S1. The S1 are then abruptly given new reference-signals and accelerate the arm downward as requested. S cannot eliminate the return swing at the beginning of the response – if he could, he might be subject to instability. (Powers, Clark, and McFarland, 1960b, p. 312)

1

A spreadsheet model of a working three level control hierarchy with five control systems at each level is available for download at www.mindreadings.com/ControlDemo/SpreadsheetHierarchy.zip.

72

Exploring the Hierarchy

This simple demonstration reveals two important characteristics of hierarchical control. The first, which is alluded to at the end of the quote, is that lower level systems must act faster than higher level ones in order for hierarchical control to work properly (without instability). In this case, the lower level system controlling the position of the arm corrects for the disturbance to arm position produced by the signal push before the higher level system can correct for the disturbance to the variable it controls – the relationship between the push signal and the position of the arm. If the higher level system could correct for the disturbance to the variable it controls before the lower level system corrects for the disturbance to the variable it controls, there could be instability in the form of uncontrolled oscillation of the arm. The fact that lower level systems must act faster than higher level systems has proved to be the basis for the development of an important tool that is used to study the control hierarchy – measures of the relative speed of operation of the control systems. The other characteristic of hierarchical control revealed by the reaction time demonstration described above is that the variables controlled by lower level systems are simpler than those controlled by higher level ones. The lower level system is controlling a perception of the position of the arm, at first keeping that perception in the state “straight out in front.” The higher level system is controlling a perception that is a combination of the perception of arm position and of the state of the signal push. It is a perception of a contingent relationship between those two perceptions. The higher level system is controlling for keeping that perception in a state that could be described as: if (no signal push) then (keep the arm out in front) else (move the arm down). Powers, Clark, and McFarland (1960b) describe several other portable demonstrations of the nested operation of control systems at progressively higher levels of the human control hierarchy. For example, they describe a demonstration of the operation of the movement control systems within the latent period of still higher level systems that control the perception of a pattern of movement. This is done by asking S to use his index finger to track E’s index finger while it moves in a circular pattern. Once S is tracking E’s finger smoothly and accurately, E suddenly stops the movement. It takes S about 500 ms (1/2 s) to stop the movement and return his finger back into alignment with E’s. During that 500 ms period, the movement control systems are keeping S’s finger moving as specified by the reference set by the pattern control systems. This is reflected in the fact that S’s finger continues to move along the arc of a circle before the movement can be stopped. So, again, the lower level systems (in this case the movement

5.2  Hierarchical Nesting of Control Systems

73

control systems) continue to operate within the latency period of the higher level system (in this case, the movement pattern control systems). And the type of perception controlled by the higher level system – the perception of circular movement – can be seen as being derived from a combination of perceptions controlled by lower level systems – the perception of arm position and arm movement (change in position). 5.2.2  Laboratory Tests of Hierarchical Control The systems demonstrated by the portable demonstrator were referred to as S1, S2, etc. to indicate their relative rather than their absolute level in a hierarchy. That is, S1 is at a lower level than S2 and S2 is at a lower level than S3. But S1 is not necessarily at the lowest level of the hierarchy, S2 is not necessarily at the next level up, and so on. There may, for example, be systems below S1 and between S1 and S2. In order to get a more precise picture of the absolute levels of different control systems in a hierarchy, it is necessary to get accurate measures of the relative timing of the operation of these systems. This kind of research requires measuring the speed of operation of a control system using laboratory tests. One approach to getting such measures was described by Marken and Powers (1989a). They had human participants do a pursuit tracking task which involved moving a mouse in order to keep a cursor aligned with a moving target. At unpredictable points during the task the polarity of the connection between the mouse and cursor was reversed; left–right movements of the mouse that had produced corresponding left–right movements of the cursor suddenly produced right–left movements of the cursor. Even though the participants knew they were coming, the polarity reversals resulted in a “runaway” condition where the cursor was “pushed away” from the target at an exponentially increasing rate as the participants continued to correct for error (deviation of the cursor from target) by moving the mouse in what was now the wrong direction. After every polarity reversal, the runaway period lasted almost exactly 500 ms after which there was an immediate change in the direction of the mouse relative to the cursor so that the cursor was again tracking the target. The result was seen as evidence that the system controlling the distance between the cursor and the target is nested within a higher level control system that is controlling for the correct polarity relationship between mouse and cursor movement. This higher level system takes ½ s to correct the polarity after the reversal occurs. During that time, the system controlling the distance between the cursor and the target continues to operate “as usual,” which,

74

Exploring the Hierarchy

due to the polarity reversal, results in a positive feedback situation, where error (the perceived distance between the cursor and the target) drives outputs that increase error. This positive feedback situation lasts ½ s at which point the higher level system is able to change the reference for the lower level system controlling the perception of the relationship between the mouse and cursor movement in a lower level system. Another approach to demonstrating hierarchically nested control systems is through the use of modeling. An example of this approach used a hierarchical control model to account for the coordinated movement behavior in an experiment conducted by Mechsner et  al. (2001). The experimental situation is shown in Figure 5.2. Participants were asked to keep two flags circling in symmetry (0 phase difference between flag movements) at an increasing speed by appropriately coordinating the movement of two handles that were hidden under a display table. Participants were able to produce this result despite the fact that the two handles did not have equivalent effects on the rotation of the corresponding flags. While there was a 1:1 relationship between left handle rotation and left flag rotation, there was only a 0.75:1 relationship between right handle rotation and right flag rotation. Therefore, in order to make the flags circle in symmetry, the handles had to circle asymmetrically (a 90° phase difference between circling handle movements). Mechsner et al. concluded that participants were able to perform this task by controlling a perception of symmetrical rotation rather than by coordinating the asymmetrical handle movements that produced this symmetry. This conclusion seemed consistent with the PCT view of

Figure 5.2  Setup of the coordination experiment.

5.2  Hierarchical Nesting of Control Systems

75

Phase Control AL

AR

p (angle)

r-p

AL - AR

r=0

Speed Control HL

(SL + SR)/2

HR

HR Speed

p (speed)

r = 1.5 r-p

r-p

HR Acc

HL Speed

r-p

HL Acc

Figure 5.3  Two-level model of coordinated movement.

purposeful behavior. This was confirmed by the fact that a two-level PCT model could account for the basic findings of the experiment. The model is shown in Figure 5.3. There are two systems at the top level (level 2) of the model; one system controls a perception of the speed of movement of the handles and the other controls a perception of the phase (angular) difference between the movements of the flags. There are two lower level (level 1) systems that control the torque exerted by the hands on each handle; one system controls a perception of the torque exerted by the right hand on the right paddle and the other controls the torque exerted by the left hand on the left paddle. These torque control systems are nested within the higher level speed and phase control systems. The higher level systems use the force control systems to control their own perceptions. They do this by varying the references sent to the two torque control systems; the output of the speed control system increases the reference to both torque control systems in order to perceive the desired increase in the average speed of the left, SL , and right, SR, flags, (SL + SR)/2; the output of the phase control system increases the reference to the left-hand target control system and decreases the speed to the right-hand torque control system in order to keep the perception of the phase difference between the angles of the left, AL , and right, AR, flags (AL − AR) equal to 0.2 2

The model can be seen in operation at www.mindreadings.com/ControlDemo/Coordination.html.

76

Exploring the Hierarchy

5.3  Types of Variable Controlled at Each Level of the Hierarchy The polarity reversal and coordinated movement studies are examples of research aimed at understanding the nested operation of hierarchically related control systems. But what we also want to know about hierarchical relationships between control systems are the types of perceptions that are controlled at each level of the hierarchy. The different types of perceptions organisms control presumably reflect the different types of purposeful behaviors organisms are seen to perform. So an important aspect of the PCT model is its proposals about what types of perceptions are controlled at each level of the control hierarchy. We will go over these proposals in some detail since understanding them is an essential prerequisite to doing research aimed at testing the PCT model of purposeful behavior. 5.3.1  The Hypothetical Levels of Perceptual Control Powers (1973b) proposed that the purposeful behavior of organisms is organized around the control of nine different types of perceptual variables, each type controlled at a different level of the hierarchy.3 A description of this proposal is shown in Table 5.1. The first column in the table gives the numerical level in the hierarchy at which each type of perceptual variable is controlled, with level 1 being the lowest and level 9 the highest. The next column gives a one-word description of the type of perceptual variable controlled by systems at that level of the hierarchy. The next three columns give examples of each type of perceptual variable in three different domains of behavior: music, painting, and athletics. The perceptual variables controlled at the lowest level of the hierarchy are called intensity perceptions, which means that they are a perception of the magnitude of stimulation at the sensory receptors. In music, this is a perception of loudness, which is proportional to the sound energy at the sensory receptors in the basilar membrane. You are controlling this perception when you adjust the volume of your speakers. In painting, an intensity perception is brightness, which is proportional to the light energy reflected from the painting. You control this perception when you adjust 3

The number of levels in Powers’s proposed hierarchy of control systems has changed somewhat over time. The nine levels in Table 5.1 are those proposed in Behavior: The Control of Perception (Powers, 1973b). The number of levels was increased to ten due to the addition of a “category” level (Powers, 1979c). Since the levels are hypotheses to be tested by future research, there is no “correct” number of levels in the hierarchy. Since the aim of Table 5.1 is simply to illustrate the basic form of Powers’s hypothesis about levels of control, only the original nine levels proposed in Behavior: The Control of Perception are included.

5.3  Variable Types by Level

77

Table 5.1  Possible levels in a hierarchy of control systems Behavioral domain Level

Perceptual type

Music

Painting

Athletics

9

System

Performance

Movement

Team

8

Principle

Interpretation

Manner

Style

7

Program

Execution

Method

Game plan

6

Relationship

Orchestration

Layout

Play

5

Sequence

Phrase

Signature

Dunk Shot

4

Transition

Dynamics

Gradient

Movement

3

Configuration

Chord

Shape

Stance

2

Sensation

Pitch

Color

Effort

1

Intensity

Loudness

Brightness

Tension

the light level in the room where the painting is viewed. In athletics, an intensity perception is the tension exerted at a joint, which is proportional to the contraction of the muscle attached to the joint. You control this perception by varying the contraction of your muscles. The perceptions at level 1 of the hierarchy are directly proportional to the environmental variables that stimulate the sensory receptors. So the perceptions at level 1 are assumed to be the only direct contact organisms have with their environment (also known as the “real world”). From level 2 on up, perceptions are assumed to be “constructed” from perceptions below them. The level 2 sensation perceptions – pitch, color, and effort – are constructed from combinations of level 1 intensity perceptions: pitch results from a combination of intensities at different locations on the basilar membrane, color results from a combination of intensities at different cone cells in the retina, and effort results from a combination of intensities at different joints in the limbs. The configuration perceptions at level 3 are assumed to be constructed from the lower level sensation and/or intensity perceptions. A configuration perception is a pattern or arrangement of lower level perceptions. An example in music is a chord, which is an arrangement of different pitch

78

Exploring the Hierarchy

perceptions; an example in painting is a shape, which is an arrangement of different color perceptions; and an example in athletics is a stance, such as that taken by a batter at the plate, which is an arrangement of perceptions of effort exerted at different points in the body. At level 4, we have transition perceptions, which are changes in lower level perceptions over time or space. One kind of transition perception in music is a dynamic change in loudness, such as a crescendo; a transition perception in painting is a gradient of color, such as the transition from dark to light blue to show a sunrise; and a transition perception in athletics is moving from one body posture (stance) to another, as when a lineman moves from a standing posture to a three-point stance. The perceptions controlled at level 5 are called event or sequence perceptions. They are defined as a fixed temporal or spatial arrangement of lower level perceptions. An example in music is a phrase or melody, which is a fixed succession of pitches or chords; an example in painting is an object, which is a fixed arrangement of shapes and shadings; and an example in athletics is the slam dunk, which is a temporal sequence of movements – run, jump, and dunk. The systems at level 6 control relationship perceptions – perceptions of the relationships between lower level perceptions. A relationship can be defined as “a regularity in the simultaneous space–time behavior of two or more independent lower-level perceptions” (Powers, 1973b, p. 286). Examples of spatial relationships are “next to” and “above”; examples of temporal relationship are “before” and “after.” An example of a relationship in music is “orchestration” which is a perception of the relationship between, among other things, the phrases played by the different instruments at different times; an example of a relationship in painting is the spatial layout of the different objects in the painting; and an example of a relationship in athletics is the relationship between the movements of the ends and backs in a pass play. The perceptual variables that are controlled at level 7 are program perceptions. A program perception is defined as “a network of choicepoints characterized by tests at the nodes” (Powers, 1973b, p. 286). The choice-points describe contingencies between lower level perceptions. A contingency is equivalent to an “if ” statement in a computer program. A contingency in a program perception might be something like “if the light is red then stop, else go.” Perceiving a program involves perceiving the appropriate lower level perceptions occurring at the choice-points in the program. An example of a program perception in music is the network of contingencies involved in producing the lower level perceptions that constitute execution of the performance of a piece; an example of a program

5.3  Variable Types by Level

79

perception in painting is the network of contingencies of lower level perceptions that are the method of producing the painting; and an example of a program perception in athletics is the network of contingencies that make up a game plan. Program perceptions are assumed to be controlled as the means of controlling the perceptual variables at level 8, which are perceptions of principles. Examples of principle perceptions are the heuristics used to solve problems. In music, the principles provide a template for how a piece should be interpreted in performance; in painting, the principles describe the manner of the picture’s composition; and in athletics, the principles define the style of the game plan. The control systems at the highest level of the hierarchy are assumed to control perceptions of systems, which are groups of entities, such as groups of people, who are organized around specific principles. Examples are nations, square dance clubs, and families. An example of a system perception in music is an orchestra, which is a group of people organized around many musical principles, such as playing in synchrony; a system perception in painting is an art movement, which is a group of people organized around a particular set of artistic principles, such as producing nonrepresentational art; and a system perception in athletics is a team, which is a group of people organized around a set of athletic principles such as “winning isn’t everything; it’s the only thing.” 5.3.2  Perception and Cognition It may seem strange to refer to the variables controlled by systems at the higher levels of the control hierarchy – variables like relationships, programs, principles, and system concepts – as “perceptions.” These variables seem to be part of one’s subjective, internal reality while the variables controlled by systems at the lower levels of the hierarchy – intensities, sensations, configurations, transitions, and sequences (events) – seem to reflect an objective, external reality. A higher level principle perception like “do unto others,” for example, seems to be a thought that exists inside of us while a lower level configuration perception, such as a round table top, seems to be something that exists outside of us. Therefore, it may seem more natural to think of the variables controlled by lower level systems as perceptions and those controlled by higher level systems as thoughts or cognitions. But PCT makes no such distinction. In PCT, the variables controlled at all levels of the control hierarchy – from intensities to system concepts – are called “perceptions” because they are all inside

80

Exploring the Hierarchy

the behaving system (Powers, 1973b); they are all assumed to be the states of neural signals. However, there may still be a question about what it means to control things like programs, principles, and system concepts as perceptions. A program, for example, seems like a contingent sequence of motor outputs, not the state of a perceptual input. You can see this in a delightful example of programmatic behavior described by Powers: I am looking for my damned glasses. First I go into the bedroom… I look at the dresser. I pick up a shirt and look under it. The glasses are not there, or anywhere else in the bedroom. Next the bathroom: on the bathtub? No. On the sink? No. In the wastebasket? No. On to the living room. Under the newspaper? Ah! End of program. Now back to the main program: to read the newspaper I put down in order to find my glasses. (Powers, 1973b, p. 160)

This program, which could be called “searching for glasses,” can be characterized as taking actions that are contingent on the results of prior actions. It can be carried out by any machine that can carry out a predefined set of operations or tasks – what is called a Turing Machine (Turing, 1948). The predefined set of operations is a program and a familiar example of a Turing Machine is a digital computer. The “searching for glasses” program could be written as a computer program and run on a computer. But while the computer was running that program it would not know whether it was running the right program. The computer could not correct itself if, for some reason, it started carrying out procedures that were not part of the “searching for glasses” program. But we know that people can correct themselves when they realize that they are carrying out the wrong program. We see it in ourselves when we wonder why we are going into the kitchen and realize we were running the “searching for a snack” rather than the “searching for glasses” program. We see it in others when we wish that they would “get with the program.” Once we realize we are running the wrong program, we can correct ourselves and go back to running the right one; we seem to be able to control for carrying out an intended program. In order to tell whether or not we are running an intended program, we must be able to perceive the program we are currently running and compare it to a reference specification for the program that should be running. And in order to control for carrying out the intended program, we have to be able to act to bring the perceived program into a match with the reference specification for that program when there is a discrepancy between perception and reference. What this shows is that you have to be able to perceive a variable in order to be able to control it. If we find that

5.3  Variable Types by Level

81

organisms are able to control complex, cognitive-type variables, such as relationships, programs, principles, and system concepts, it’s because they can perceive these variables. 5.3.3  Evidence for the Levels The levels of the perceptual control hierarchy described in Table 5.1 represent a hypothetical framework for understanding all the different kinds of purposeful behaviors that we see organisms doing. Simple behaviors, like lifting a finger or a foot, presumably involve the control of simple, lower level perceptions, such as intensities and sensations. More complex behaviors, like playing the piano or dancing a waltz, presumably involve the control of more complex, higher level perceptions, such as sequences and relationships. The levels also reflect a hypothesis about the perceptual processing done at different levels of the nervous system. So both behavioral and neurophysiological studies can be done to test the hierarchical control model described in Table 5.1. 5.3.4  Neurophysiological Evidence of Levels Powers (1973b) described some of the evidence for the existence of control systems in the nervous system that control perceptual variables corresponding to those proposed for levels 1–5 (intensity to sequence). Anatomical evidence suggests that level 1 intensity control systems are implemented at the level of sensory receptors and spinal reflexes of the nervous system (Powers, 1973b, pp. 82–89) and level 2 sensation control systems are implemented anatomically above that in the sensory and motor nuclei of the brain stem (Powers, 1973b, pp. 100–107). Level 3 configuration control systems are most likely implemented in the midbrain and cerebellum. Based on the classic electrical stimulation studies by the Nobel Laureate Walter Hess Powers also concluded that level 3 configuration control systems are implemented above the brain stem in the thalamus (Powers, 1973b, pp. 118–119).4 There was less experimental evidence when Powers (1973b) attempted to anatomically locate levels 4 and 5; however, he suggested that Level 4 transition control systems could be implemented above the thalamus in the secondary somatosensory 4

Although Powers (1973b) displayed great knowledge of neuroanatomy, he was not a neuroscientist and he did make a couple of mistakes. One such mistake is where he equates the sensory nuclei of the thalamus with the midbrain (p. 114). The midbrain is not part of the thalamus, and thalamic nuclei do not reside in the midbrain.

82

Exploring the Hierarchy

area of the cerebral cortex (Powers, 1973b, p. 134), and level 5 sequence control systems could be implemented in either the limbic system or motor cortex (Powers, 1973b, p. 139). More recent neurophysiological research seems to be generally consistent with these earlier findings. With clever experimentation from his own lab, Yin (2016, 2017) has outlined in greater detail and with more anatomical precision the location of the different levels of control, up to level 6 – control of relationships. He has found that configurations, transitions, as well as relationships (and possibly sequences) are controlled by different nuclei associated with the basal ganglia. The basal ganglia are a group of subcortical nuclei located below the cerebral cortex and are reciprocally connected with the brainstem, thalamus, and cortex. For example, work from Yin’s lab has shown that neurons in the substantia nigra pars reticulata (SNr) of the midbrain represent instantaneous position coordinates of the body (a configuration perception; Barter et al., 2015), and that striatal medium spiny neurons in the basal ganglia represent vector components of velocity (a transition perception; Kim et al., 2014). The SNr represents the main output nuclei of the basal ganglia and receives direct input from the striatal medium spiny neurons. Kim et al. (2019) found that the relationship between the position of a mouse’s head relative to a moving reward spout – a relationship perception – is controlled by striatal parvalbumin interneurons, which are connected to the medium spiny neurons. Finally, Hughes et al. (2019) found that control of head angle – a configuration perception – is also located anatomically below control of velocity (a transition perception) in the ventral tegmental area of the midbrain, which also receives input from the striatum. Taken together, these findings are evidence of three higher levels of control: relationship references (from the cerebral cortex) are being sent into transition comparators within the striatum; transition errors are then projected to the midbrain, which contain neural integrators that convert the transition errors into configuration references (Yin, 2017). There seems to be very little neurophysiological evidence for control systems above the relationship perception level (level 6) – the systems controlling programs, principles, and system perceptions. Most of the evidence for the existence of higher levels of perceptual control comes from behavioral experiments. 5.3.5  Behavioral Evidence for Levels An approach to obtaining behavioral evidence for the hierarchical model was described by Marken (2002b). The approach is based on the idea

5.3  Variable Types by Level

83

that higher level systems control on a slower time scale than lower order systems. Therefore, one way to test the hierarchical model is to present different types of perceptual variables at different speeds and determine the minimum rate at which each variable can be detected. This can be done using a version of the psychophysical method of adjustment, where the participant is asked to adjust a parameter of the stimulus, such as its intensity or (as in this case) its rate of presentation until it is just detectable. In one such study (Marken, 2002b, pp. 100–106), a computer program presented a sequence of single-digit numbers at two different positions on the display. The presentation positions were vertically adjacent and horizontally separated by 2 cm. The numbers were presented alternately to the two positions. Participants adjusted the rate at which the numbers occurred in each position by varying the position of a mouse controller. There were four aspects of the alternating numbers that the participants could perceive. These represented four different types of perceptual variable in the control hierarchy: (1) configuration, which is the identity of each number; (2) transition, which is the fact that the numbers moved back and forth between the two positions; (3) sequence, which is the order in which the different numbers are presented in each position; and (4) program, which is the rule that determined which number followed the next in a sequence. The test began with the numbers presented at a very high rate and participants were asked to adjust the rate of presentation until they were just able to identify each of these aspects of the alternating numbers. The results were consistent with the order of levels of perceptual types in the PCT hierarchy (Table 5.1). Configurations (numbers), which are thought to be at the lowest level of the hierarchy for variables included in this research, were perceived at the fastest rate (20/s); transitions (back and forth movements), which are thought to be at the level above configuration, were perceived at a slower rate (7/s); sequences, which are thought to be at the level above transitions, were perceived at a still slower rate (4/s) and programs (the rule that generated the number sequence) were perceived at the slowest rate (0.25/s). These rate measures varied somewhat, but there was no overlap across measures so the order of rates was always consistent with the assumed order of the levels of these perceptual types in the hierarchy, with configuration perceptions at the lowest level and program perceptions at the highest. These results are consistent with the idea that these different types of perceptual variable are computed at different levels of the nervous system and that the order of levels is consistent with that shown in Table 5.1. But there is still the question of whether these perceptual variables are

84

Exploring the Hierarchy

controlled at these different levels. A basic assumption of the hierarchical control model is that lower level systems operate more quickly than higher level ones. As noted earlier, this is assumed to be a basic requirement for stable control at all levels of the hierarchy; lower level systems must bring the perceptions they control to the reference states specified by higher level control systems before the higher level systems change those references or the higher level systems will start to oscillate, resulting in loss of control. One way to test this assumption uses a variation of the method of adjustment described earlier (Marken, Mansell, and Khatib, 2013). Again, visual stimuli that could be perceived in different ways were presented to participants at different rates. In this case, the stimuli were objects rather than numbers and they were displayed in a circular series of positions on a computer screen, as shown in Figure 5.4. The objects, which were shapes of different sizes, could be perceived in terms of configuration (circle vs square), transition (clockwise or counterclockwise movement), or sequence (“small, medium, large” or “small, large, medium”). At random times during a trial, a disturbance would change the state of each of these perceptual aspects of the display to its opposite: the disturbance to the configuration would change the circle to a square or vice versa; the disturbance to the transition would change clockwise to counterclockwise motion or vice versa; and the disturbance to sequence would change the sequence from “small, medium, large” to “small, large, medium,” or vice versa. The participants in this experiment were asked to control each of the different perceptual aspects of the display by keeping it in one particular reference state; controlling a configuration perception meant keeping it in the state “circle,” controlling a transition perception meant keeping it in the state “clockwise movement,” and controlling a sequence perception meant keeping it in the state “small, medium, large.” In order to do this, the participant had to act, by pressing the mouse button, to compensate for

Figure 5.4  Four sequential frames of the animated display showing the objects moving clockwise and increasing in size (t0), changing shape (t0+t), changing sequence (t0+2t), and changing direction (t0+3t) (from Marken, Mansell, and Khatib, 2013, Figure 3).

5.3  Variable Types by Level

85

the randomly occurring disturbances generated by the computer. Pressing the button changed the state of the perceptual variable, returning it quickly to its reference state if the button was pressed immediately after the disturbance. How well each perception was being controlled was measured in terms of the proportion of a trial that the perception was kept in the reference state (“on target”). The results of this experiment are shown in Figure 5.5. The ability to control each perception is shown as a function of the presentation rate, which is the rate of presentation of the different shapes in the animation frames. The results are consistent with the hypothesis that systems controlling lower level perceptual variables operate much more quickly than those controlling higher level perceptual variables. Control of configuration was near its maximum value when the presentation rate was 10/s. Control of transition didn’t reach that level of control until the presentation rate was just 4/s. And control of sequence only started to approach that level of control when the display rate was 2.2/s. These results also show that the constraint on the speed of operation of control systems at different levels is on the input, not the output, side of the control system. This is demonstrated by the fact that the same simple output – a press of the mouse button – was used to control all three perceptions that were controlled in this study. This suggests that the main

Control (proportion of trial on target)

Control as a Function of Presentation Rate 1.00

0.80

0.60 Configuration

0.40

Transition Sequence

0.20

50

100

200

450

Presentation Rate (msec/frame)

Figure 5.5  Controllability of different perceptual variables as a function of rate for presentation (adapted from Marken, Mansell, and Khatib, 2013, Figure 4).

86

Exploring the Hierarchy

limitations on our ability to rapidly produce certain behaviors are in our ability to rapidly generate the perceptions to be controlled rather than the outputs that control these perceptions. For example, we can play a trill on the piano more rapidly than we can play the theme to Beethoven’s Fifth (da da da daaah) with the same two fingers because the trill involves control of a lower level perception (transition) than the one controlled when playing the Fifth (sequence). 5.3.6  Testing for Perceptual Types The example of playing a trill versus playing the theme of Beethoven’s Fifth on the piano illustrates what is meant by the concept of perceptual types. The trill and “da da da daaah” perceptions are equivalent to the “clockwise movement” and “small, medium, large” perceptions that were controlled in the Marken, Mansell, and Khatib (2013) experiment. Both the trill and “clockwise movement” are transition-type perceptions; both “da da da daaah” and “small, medium, large” are sequence-type perceptions. The trill differs from “clockwise movement” and “da da da daaah” differs from “small, medium, large” only in terms of the lower level perceptions of which they are composed. The trill and “da da da daaah” perceptions are composed of lower level auditory perceptions and the “clockwise movement” and “small, medium, large” perceptions are composed of lower level visual perceptions. So a perceptual type can be defined as a set of perceptions that are the same function of different kinds of lower level perceptions. The concept of a perceptual type is reflected in Table 5.1 by the examples of perceptions controlled in music, painting, and athletics that correspond to the types of perceptions that are presumed to be controlled at each level of the hierarchy. A chord in music, a shape in painting, and a stance in athletics are all configuration-type perceptions. They are configuration perceptions in the sense that they are arrangements of lower level perceptions from different sense modalities; a chord is an arrangement of auditory perceptions, a shape is an arrangement of visual perceptions, and a stance is an arrangement of proprioceptive perceptions. But the idea that chords, shapes, and stances are the same kind of perception could be just a matter of opinion. What is needed is an objective way to determine whether different perceptual variables really are of the same type. And again the objective way to do this is based on measures of the speed at which a perceptual variable can be controlled.

5.4  Summary

87

Studies of the perception of auditory trills indicate that the fastest rate at which “movement” from one tone to another can be perceived is 10 tones/s (Miller and Heise, 1950), about the same as the fastest rate at which movement of a shape from one position to another can be perceived. Studies of the perception of tone sequences show that the fastest rate at which such sequences can be perceived is 4 tones/s (Warren et al., 1969), which is about the same as the fastest rate at which a visual sequence can be perceived. Since playing a trill or tonal sequence on the piano also involves the production of proprioceptive perceptions – perceptions of one’s own finger movements – the same rate limits should also hold for the ability to perceive proprioceptive transitions (trills) and sequences (key press sequences). And this seems to be the case (Rosenbaum, 1987). Research on the types of perceptual variables organisms control opens up the possibility of developing a new kind of “mental chronometry,” one that uses reaction time to map out the hierarchy of types of perceptual variables that an organism controls. Mental chronometry was originally developed by Donders (1868, reprinted in 1969) as a way to measure human mental processes using reaction time. The use of reaction time to study mental processes has evolved into a sophisticated staple of research in cognitive psychology (e.g., Neisser, 1967; Sternberg, 1969; Shepard and Metzler, 1971). But this research is done in the context of an open-loop model of behavior. The hierarchical PCT model provides a new organizing framework for understanding mental processes in terms of a closed-loop model of behavior. In the closed-loop model, reaction time is reflected in the time scale on which different types of perceptual variables can be controlled. Research aimed at measuring these time scales could provide an objective basis for classifying controlled variables into types to see whether those types correspond to those listed in Table 5.1. Perceptual variables would be considered to be of the same type not only if they seem to be the same kind of variable but if they are also controlled on the same time scale.

5.4  Summary Perceptual Control Theory assumes that purposeful behavior is produced by a hierarchy of control systems, with control systems at each level of the hierarchy controlling different types of perceptual variables. This chapter described the two main lines of research that have been aimed at exploring different aspects of this hierarchy. One line of research is aimed at testing the proposed nested relationship between systems in the hierarchy. The

88

Exploring the Hierarchy

other is aimed at testing Powers’ proposal regarding the types of perceptual variables controlled by control systems at each of the different levels of the hierarchy. Both lines of research explore the hierarchy using reactiontime-type experiments. This is based on the fact that a basic stability requirement for control systems operating at different levels of the hierarchy is that lower level systems operate on a faster time scale than higher order systems. Reaction-time-type experiments can be used to evaluate nested relationships between control systems and as a basis for classifying the variables controlled at different levels of the hierarchy into types.

6

Learning

The control hierarchy doesn’t work perfectly when it comes “right out of the box”; organisms have to learn to control the perceptions involved in carrying out various behaviors, sometimes even those behaviors that are essential to their survival. Soon after it is born, for example, a foal has to learn to control the perceptions involved in standing on its own four feet in order to suckle from its mother or it will starve. Baby humans have to learn to stand too, but it is not quite as urgent as it is with baby horses since human mothers help out with the nursing soon after birth. But it is clear that controlling has to be learned and a complete theory of purposeful behavior has to explain how this learning occurs.

6.1  Reorganization of Control Systems Perceptual Control Theory (PCT) views learning as a process of developing the neural structures that become the systems that control the perceptual variables that the organism must be able to control in order to survive. The process is called reorganization because it is presumed to involve reorganizing the existing control structures so that they effectively control the required perceptions. The existing control structures are the hierarchical arrangement of control systems described in Chapter 5. What get reorganized are the neural networks that make up the main functional components of each control system in the hierarchy – the input, comparator, and output functions – as well as the connections between those systems. What does the reorganizing is the reorganization system – a “meta” control system whose purpose is to keep the hierarchy of control systems functioning properly. The relationship between the reorganization system and the existing hierarchy of control systems is shown in Figure 6.1. The reorganization system is the control system on the left side of the diagram in Figure 6.1. The “environment” of this control system is the 89

90

Learning

intrinsic state of the organism, which consists of the physiological variables that must be kept under control in order for the organism to survive. This would include intrinsic variables that are indications of physical health, such as measures of blood sugar and oxygen levels, but it can also include variables that are indications of psychological health, such as measures of how well systems in the control hierarchy are functioning (Marken and Carey, 2014). The reference specifications for the intrinsic variables are provided by the genes (the genetic source in Figure 6.1). Any discrepancies between intrinsic variables and the genetically determined reference specifications for the states of these variables are intrinsic errors.

Reorganizaon System

Learned Hierarchy of Control

GENETIC SOURCE INTRINSIC REFERENCE SIGNALS INTRINSIC PERCEPTUAL SIGNALS

INTRINSIC ERROR SIGNALS

COMPARATOR Homeostac control system

INPUT FUNCTION

OUTPUT FUNCTION

ORGANIZATION ALTERING EFFECTS

NEURAL OR CHEMICAL SIGNALS

INTRINSIC STATE SENSORY EFFECTS

BEHAVIORAL ACTS

PHYSIOLOGICAL EFFECTS

ENVIRONMENT Figure 6.1  The reorganization system in relationship to the hierarchy of control systems.

6.2  Evidence of Reorganization

91

These errors drive outputs that reorganize the control hierarchy by making changes to the input and output functions of individual control systems as well as to connections between higher and lower level control systems. These changes result in changes in the behavior of the control hierarchy which affect the intrinsic state of the organism. When these changes in behavior result in the elimination of the intrinsic error, the reorganization process stops; the organism is now behaving in a way that maintains its intrinsic variables in their genetically determined reference states. The reorganization system makes random changes to the control hierarchy because, like any true learning system, it doesn’t already know how to make the changes that will result in behavior that eliminates intrinsic error. If it did then there would be no learning required since removing the error would be something the reorganization system already knows how to do (Petrie, 1981). But the random change process used by the reorganization system is of a particular sort that gives it a purpose; the purpose of eliminating intrinsic error. This “purposeful” random change process has been dubbed E. coli reorganization after the system used by the Escherichia coli bacterium to navigate to food sources (Koshland, 1980). E. coli reorganization is a biased random-walk process that turns out to be highly efficient (Marken and Powers, 1989b). The idea is that the reorganization system works by varying the rate at which random changes are made to parameters of the control systems in the control hierarchy. When intrinsic error is increasing, these changes are made at a higher rate than when intrinsic error is decreasing. As intrinsic error approaches zero the rate at which changes are made also goes to zero (Powers, 2005a, pp. 107–126). When intrinsic error is zero, the control systems in the hierarchy are controlling in a way that is keeping the intrinsic variables matching their references; everything is now working properly so no change in the operation of these control systems is needed or wanted.

6.2  Evidence of Reorganization We are getting a picture of the operation of the reorganization system when we see the somewhat random-looking “trial and error” behavior of an organism trying to learn a new skill. But the picture we get is somewhat smudged. This is because any skill involves controlling many different perceptual variables at the same time. So learning a skill must involve making random changes in the parameters of many different control systems simultaneously. Some of these are changes to input functions, resulting in changes in the perceptual variables the system is (or should be) controlling. Others are changes to output functions, resulting in

92

Learning

changes in the way the system sets the reference signals going to lower order systems or, at the lowest level in the hierarchy, changes in the way the system affects the aspect of the environment that corresponds to the controlled variable. According to the reorganization theory of learning, the random trial and error behavior we see when an organism is learning a new skill is actually a mix of random changes being made to the components of many different control systems at many different levels of the control hierarchy. The random trial and error behavior seen when a foal is learning to stand on four legs, for example, probably involves random changes in the components of control systems that control muscle tensions, force vectors, joint angles, visual orientation, and balance, among other things. As the systems controlling these perceptions become more effective, the rate of random change decreases and the foal’s jerky, trial and error efforts give way to a stable stance and gait. The goal of research on reorganization is to see the workings of the reorganization system through the smudginess of overt trial and error behavior. A nice example of this kind of research was done by Robertson and Glines (1985) using a learning task designed by Powers (1985). The task was implemented as a simple computer game where the participant’s goal was to “beat the computer” by learning to press keys on the keyboard in a way that prevents the computer from getting points. The computer gets points by sequentially cycling a star through four different boxes on the display screen at a fixed rate. Points are added to the computer’s score as long as the star is showing in one of the boxes. The participant can interrupt the cycle by pressing the key associated with the box in which the star is currently located. When the correct key is pressed, the star disappears and the computer’s accumulation of points stops until the star appears in the next box in the sequence. So the participant can slow the computer’s accumulation of points by reacting quickly with the correct keypress as soon as the star appears in a box; the faster the reaction time, the slower the accumulation of points by the computer. The participant can take points away from the computer and win the game by learning to press the correct key before the star appears in the next box in the cycle. This game was designed so that participants had to learn to play in a sequence of different levels of mastery, each level characterized by the ability to control a different kind of perceptual variable. It was also designed so that these different levels of mastery involved control of perceptual variables that were hierarchically related. This was done by having “… the perceptual variable to be controlled at each level subsume the variable

6.2  Evidence of Reorganization

93

previously brought under control at the level below” (Robertson and Glines, 1985, p. 56). Mastery was measured in terms of the participant’s ability to slow or reverse the computer’s accumulation of points. The lowest level of mastery required learning to control a perception of the correct key to press when the star is in each box. The next level of mastery required learning to control a perception of the correct sequence of keypresses, one that corresponds to the sequence of boxes in which the star appears. The highest level of mastery required learning to control a perception of a temporal relationship between the correct sequence of keypresses and the computer-generated sequence of stars; the keypress sequence had to anticipate the star sequence. The hypothesis tested in this experiment was that in order to win the game the participant would have to master control of each of these perceptual variables. And because these variables are hierarchically nested, control of these variables would have to be mastered in order from the lowest to the highest level variable in the hierarchy. Participants would have to learn to control for making the correct keypresses before they could learn to control the correct sequence of keypresses; and they would have to learn to control for making the correct sequence of keypresses before they could learn to control for having this sequence of keypresses anticipate the sequence of stars displayed by the computer. And most importantly, it was expected that there would be a period of reorganization prior to mastery of each level of control. This period of reorganization was expected to show up as a deterioration of performance in the form of somewhat randomappearing keypresses. Only about 50 percent of the participants tested in this study were able to win the game by bringing the computer’s point count to zero. That is, only half the participant learned to control the highest level variable – keypress sequence correctly anticipating the display sequence. The result for one of the participants who did learn to control the highest level perception and win the game is shown in Figure 6.2. The graph shows the time between keypresses (inter-response time, IRT) during each display cycle from the start of the game to the point where it was won. What we see are periods of variable IRTs alternating with periods of fairly constant ones. The periods of variable IRTs presumably reflect periods of reorganization where the participant is trying to figure out what perceptions to control and how to control them in order to win the game. The periods of nearly constant IRTs, which the authors refer to as plateaus, presumably reflect points where the participant has learned to control a perception that provides some control over the computer’s point count.

94

Learning Reorganizing III

II

IV

Reorganizing

Inter-response Time

I

Time

Figure 6.2  Reorganization periods followed by stability (adapted from Robertson and Glines, 1985, Figure 1).

These results are consistent with the hypothesis of the experiment: in order to win the game, the participant had to learn to control at least three different perceptions, one after the other. The periods when these perceptions were learned correspond to the plateaus indicated by Roman numerals. The game starts with a period of highly variable IRTs, which indicates that reorganization is going on – random changes in action are occurring at a high rate. This reorganization period is followed by a plateau (period I) where the rate of reorganization has decreased to nearly zero. During this plateau, the participant is able to control for correctly associating keypresses with the location of the star. After plateau I, there is another period of reorganization, which is followed by plateau II where the participant has learned to control the correct keypress sequence. After plateau period II, there is another period of reorganization which the authors see as being interrupted by a brief period of control that they call plateau III. But it is not clear what perceptual variable is being controlled during this plateau. It seems likely that the entire period from the end of plateau II to the beginning of what the researchers call plateau IV is a period of reorganization. The variations in IRTs at the end of this period (and prior to plateau IV) are not random, but they are changing; they are getting shorter which suggests that reorganization is hot on the trail of the perceptual variable that, when controlled, will allow the participant to win the game by making keypresses that correctly anticipate the display sequence. The Robertson and Glines experiment was designed to reveal one hypothetical characteristic of the reorganization system: the process of finding

6.2  Evidence of Reorganization

95

the correct lower level perceptions to control in order to control a single higher level perception – the perception of beating the computer. The experiment was set up as a game that could be won only by controlling a particular perception: a keypress sequence that correctly anticipates the display sequence. Control of this perception presumably requires the control of two nested perceptions: (1) the correct association between keypresses and display items and (2) the correct keypress sequence. Since control of the higher level perception depended on the ability to control the lower level perceptions, control of the lower level perceptions had to be learned before it was possible to control the higher level one. The results in Figure 6.2 show that this is, indeed, what happened. The constant slow IRTs of plateau I show that the participant first learns to control a perception of the correct association between keypresses and display items; the slightly faster IRTs of plateau II show that the participant next learned the correct keypress sequence; and the very fast IRTs of plateau IV (which is probably actually plateau III) shows that the participant finally has learned to control the perception of the keypress sequence correctly anticipating the display sequence, which makes it possible for the participant to control for beating the computer. The Robertson and Glines study provides a nice, high-level view of learning a hierarchy of skills via a random reorganization process. Variations of this type of study could be done to get a more detailed picture of how reorganization progresses. For example, studies could be done to determine whether random variations are truly random and whether there is a difference in the efficiency of reorganization depending on the level of the variable to be learned. The plateaus in Figure 6.2 are interesting in themselves because they show that, in this particular learning situation, the random change aspect of reorganization behaves pretty much as expected: the process stops or slows considerably once it has hit on a change that reduces the rate of increase in error – the error being the increase in the points gained by the computer. There is some variation in the IRTs in each plateau but it is likely that whatever changes occur during this period just make things worse so the system is returned to its prechange state. The fact that the reorganizing periods – the periods when the IRTs become more variable again – always end up with a system that controls a new perceptual variable of which the previously learned variable is a component suggests that each reorganization period was working on developing a higher level control system. What seems to be happening in this learning situation is a phenomenon called “going up a level” that has been observed in psychotherapeutic situations (Carey, 2006).

96

Learning

6.3  Evidence of Reorganization in Psychotherapy In psychotherapy, “going up a level” occurs in the context of a client’s attempt to resolve internal conflict. An internal conflict exists when two or more control systems are controlling the same or a similar perceptual variable relative to two different reference specifications. For example, a client with an eating disorder is probably in a conflict over a perception of the amount of food to eat when hungry. At least two control systems in the client want to keep this perception in two different reference states. One system – perhaps the one controlling for the perception of taste – wants to eat a lot of food when hungry; another system – perhaps the one controlling for the perception of thinness – wants to eat little or nothing when hungry. The systems controlling for taste and thinness are trying to get the reference for the same variable – amount of food eaten – into two different states. The result is that there is a loss of control of how much is eaten. Behaviorally, this is seen as an eating disorder such as anorexia (avoiding eating at all) or bulimia (oscillating between eating and throwing up; binging and purging). The client’s loss of control due to conflict is analogous to the Robertson and Glines study participant’s lack of control due to ignorance of how to win the game. In both cases, there is a desire to be in control – of how much to eat or of how many points the computer has – there is just a lack of the knowledge of what to do to have such control. And in both cases, the way to get that knowledge is by a process of reorganization that involves “going up a level” (Carey, 2008). In the Robertson and Glines study, going “up a level” meant finding the higher level perceptual variable that, when controlled, improves the ability to control the computer’s point count. That variable was the anticipatory relationship between the keypress and display sequence. In the case of therapy for an eating disorder, “going up a level” means finding a higher level variable that, when controlled, improves the ability to control both thinness and how much one eats when hungry. Perhaps that higher level variable is something like a perception of one’s relationship to others – a system concept perception – which can be controlled by means other than by being very thin. The similarity of the processes involved in learning and psychotherapy suggests that research along the lines of the Robertson and Glines experiment should be able to tell us much about the reorganization process that is common to both, in particular the conditions under which reorganization begins and ends (Powers, 1985). The results of such research could be of considerable practical value to the extent that they could help

6.4  An Ethological Approach to Studying Reorganization

97

practitioners know how to establish the conditions that are propitious to a successful start and end of therapy.

6.4  An Ethological Approach to Studying Reorganization Reorganization has also been studied by observing the development of apes and human infants in natural settings, an approach to research used in the branch of biology called ethology. Of particular interest is the work of Frans Plooij and his wife Hetty van de Rijt-Plooij because it was carried out with an understanding that the behavior that they were observing was a control process (Plooij and van de Rijt-Plooij, 1990). Because the Plooijs were familiar with the hierarchical PCT model of purposeful behavior, their research was oriented toward finding out what kinds of perceptual variables infants were able to control at different points in their development. The Plooijs conducted much of their research in natural settings – indeed, Frans’s initial studies of the development of infant chimpanzees were done in Gombe National Park, Tanzania, under the supervision of Jane Goodall – so the methods they used to do research on purpose – in particular, the TCV – had to be adapted to a nonlaboratory environment where it was impossible to manipulate variables under controlled conditions. The solution was to use an observational version of the TCV. Plooij (1987) describes three ways to test for controlled variables using observation alone: (1) look for resistance to naturally occurring disturbances, (2) look at the speed of control, and (3) look for consistency (low variability) of a result of action. Using these methods, the Plooijs were able to identify eight different types of perceptions controlled by infant chimps, which correspond to the first eight levels of the proposed control hierarchy shown in Table 5.1 (Plooij, 1987, pp. 70–73). Whether these perceptual types have been correctly categorized is a matter for further research. But what the Plooij’s discovered that is relevant to reorganization is that an infant chimp’s ability to control different types of perception develops in a distinct developmental sequence. For example, the ability to control what they identified as intensity- and sensation-type perceptions develops shortly after birth; the ability to control configurations develops next, at around 2 months, followed by the ability to control transitions at around 3–4 months, sequences at 5 months, and so on. As each of these capabilities comes “on line,” there is a brief period of instability – what the Plooijs call regression periods – during which there is a deterioration of the infant’s current control capabilities. After

98

Learning

each regression period there is a period of stability when the infant can control its world in terms of the new type of perceptual variable as well as in terms of perceptual variables that have already been learned. For example, the first time a baby chimp is able to scratch itself is at around 1 ½ to 2 months. The infant can do this because it is able to form its hand into the configurations needed to produce scratching. But this scratching is carried out in a very “wooden” way; the infant is unable to transition smoothly from one hand configuration to another, suggesting that it is not yet able to control transition-type perceptions. This wooden way of scratching continues for about a month until the infant goes through a brief regression period during which the scratching movements are performed even more poorly. When the infant emerges from this regression period, it is making smooth scratching movements, indicating that the infant has added the ability to control transition perceptions to its behavioral repertoire. The regression periods that are seen in the development of baby chimps are also seen in the development of baby humans (Rijt-Plooij and Plooij, 1992; Plooij, 2013). And in both cases, they resemble the periods of random reorganization that occur between the plateaus in the Robertson and Glines study. Indeed, the trajectory of development of control skills in baby chimps and humans bears a striking resemblance to the trajectory of development of control skills in the adult humans in the Robertson and Glines study. The trajectory of development in infant chimps and humans involves periods of stability alternating with periods of instability. The periods of stability reflect skilled control of a particular type of perceptual variable; the periods of instability reflect a process of reorganization, when the infant is developing the ability to control a higher level perceptual variable so that it is able to control in a new and more effective way. The fact that the controlling done in the stable periods following instability is more skillful than it was in the stable periods preceding instability suggests that the periods of instability are, indeed, periods of reorganization. And in the case of both the infant chimps and the adult human gamers, the result of the reorganization is apparently the ability to control a new perceptual variable that makes it possible to control another perceptual variable more effectively. In the case of the baby chimp, reorganization makes it possible to scratch an itch more effectively. In the case of the gamer, reorganization makes it possible to compete with the computer more effectively. The ethological approach to the study of reorganization provides a very promising avenue for research aimed at testing the PCT model of purposeful behavior. Indeed, the studies done to date have provided

6.5  Consciousness: The Control of Reorganization

99

information about human development that has proven to be of great practical value to parents (Vanderijt and Plooij, 2003). For one thing, these studies tell parents to expect periods of behavioral instability and when to expect them. The periods of instability show up as times when the infant is fussier than usual and prone to tantrums and illness. Parents who know that these periods are part of the normal process of development – indeed, that they are a necessary part of development – are able to deal with them with greater equanimity than are parents who don’t.

6.5  Consciousness: The Control of Reorganization With the addition of the reorganization system to the hierarchy of control systems, the PCT model of the purposeful behavior of living systems is nearly complete. One important psychological phenomenon that is still missing is consciousness, something we know should be in a model of the behavior of living control systems mainly because we are living control systems ourselves. But if we put it in the model, where should we put it? The current hypothesis is that consciousness is associated with the system that is involved in learning new or refining existing skills: the reorganizing system. Consciousness fits into the reorganization process rather nicely when we think of it as made up of two components: awareness and volition. Awareness is the “input” side of consciousness and can be defined as the “perception of perceiving” (Powers, 1973b, p. 199). When defined in this way, awareness can be seen as being related to the aspect of reorganization that monitors the state of the control hierarchy. The perceptual variables in that hierarchy are the objects of awareness; they are the perceptions – the “perceiving” – that awareness perceives. Volition, then, is the “output” side of consciousness and can be defined as the “injection of arbitrary signals … into the hierarchy of control systems by the reorganization system” (Powers, 1973b, p. 288). When defined this way, volition can be seen as being related to the aspect of reorganization that makes random (arbitrary) changes to the control systems that need fixing. The fact that consciousness, in the form of awareness and volition, maps so closely to functions of the reorganization system makes sense since learning new skills or perfecting existing ones seems to require consciousness. It is unlikely that the participants in the Robertson and Glines study could have learned the perceptions to control in order to win the game if they were unconscious. Indeed, consciousness maps so closely to the functions of the reorganization system that one might conclude that the reorganization system is consciousness. Adding consciousness to

100

Learning

the PCT model of learning would thus be adding nothing more than some new words – awareness and volition – to describe the function of the reorganization system. But consciousness does have a property that adds an important functionality to the reorganization system; consciousness has mobility so it can help direct reorganization to the places in the hierarchy where reorganization is needed. The awareness aspect of consciousness seems to move around our experience of the world like a searchlight moving across the sky. In PCT, the mobility of awareness is presumed to correspond to looking at the control hierarchy from different points of view, each point of view corresponding to that of a different control system in the control hierarchy. So you can be aware of your experience of reading this text from the point of view of systems that perceive the meaning of the text. Or you can move your “awareness searchlight” and look at what you are reading from the point of view of systems that perceive the spelling of words in the text. While aware of the world from one point of view, other points of view are no longer in the field of experience; while reading this text for meaning you are probably not aware of how the words are spelled, though your awareness may be dwarn to the spelling point of view by egregious misspellings, like the one earlier in this sentence. The mobility of awareness contributes an important functionality to the reorganization model of learning by making it possible to focus reorganization, in the form of arbitrary, volitional changes in control parameters, on particular control systems in the hierarchy. The assumption is that reorganization is applied only from the current conscious point of view. Indeed, reorganization is assumed to start as soon as awareness moves to that point of view. So reorganization of the system controlling for winning the game in the Robertson and Glines experiment will begin once the participant is consciously experiencing the world from the point of view of playing the game; reorganization of the system controlling for eating reasonable-sized meals when hungry will begin once the person with the eating disorder is consciously experiencing the world from the point of view of the size of the meals being eaten. This view of how consciousness contributes to the reorganization process has some support from anecdotal evidence. For example, performance of a well-learned skill, such as playing a musical instrument, is known to deteriorate when one’s awareness moves from the point of view of the desired result being produced (the music) to how that result is being produced (the actions used to make the music). Moving awareness to the point of view of how one is producing a particular result apparently kicks

6.5  Consciousness: The Control of Reorganization

101

off reorganization of control systems that are already properly organized, thus leading to a deterioration of a well-learned skill. Other evidence that the awareness aspect of consciousness is enough to initiate reorganization comes from psychotherapy. Psychotherapists of all persuasions seem to agree that before a person can solve their problem, they must become aware of it from the appropriate point of view, whether that point of view is described as recall of past trauma, awareness of unconscious desires or awareness of conflicting goals. Since, from a PCT perspective, solving psychological problems is a process of reorganization, the experience of psychotherapists is consistent with the idea that finding a solution to one’s problems – reorganization – requires awareness of the problem from the right point of view. While the anecdotal evidence suggests that awareness does initiate reorganization, this idea should be tested experimentally. Powers (1980) has suggested a way to do this using the warhorse of laboratory methods for testing PCT, the compensatory tracking task. These thoughts lead to the idea of constructing some multileveled control experiments (experiments concerning control, that is) in which the subject is encouraged, asked or underhandedly forced to concentrate on one of the hypothetical levels of perception involved, enough to drive the others out of immediate attention…. If awareness has the properties of something that can move from one point of view to another, and if it is intimately or even necessarily involved in the process of reorganization, the parameters of control ought to become variable at the level where the current point of view is located. It is not excessively difficult to monitor a few basic parameters of control, such as sensitivity, phase shift and RMS (root mean square) error on a continuous basis…. (Powers, 1980, pp. 241–242).

What Powers is suggesting is research aimed at looking for evidence of reorganization that is initiated by changes in one’s conscious point of view regarding the task being performed. If consciousness works as it is thought to in the PCT model of reorganization, then we should see changes in the temporal variability of the parameters of control – sensitivity (which is equivalent to output gain), phase shift (which is equivalent to speed of operation), and RMS error (which is a measure of how well the person is controlling) – as a function of changes in the conscious point of view of the person doing the control task. Powers proposes that such an experiment could be carried out using a two-level control task, which is most easily implemented as a compensatory tracking task. The upper level of control in this task would be control of the visual perception of the distance between the cursor and the target. The lower level would be control of the

102

Learning

proprioceptive perception of the force exerted on the output device – the mouse, joystick, etc. – that is used to control cursor–target distance. In order to continuously compute the parameters of these two control systems, it will be necessary to apply uncorrelated, time-varying disturbances to both controlled perceptions – cursor–target distance and the force required to move the output device. The continuous monitoring of a “few basic parameters” of control that would not have been “excessively difficult” back in 1980 would not be difficult at all with the computers of today. Indeed, using the computer technology of today, it should be possible to set up an experiment that allows real-time monitoring of the variability of control parameters as a person carries out Powers’ proposed consciousness test. If, indeed, awareness does initiate reorganization in the form of random changes in control parameters, then as a person shifts his/her awareness from the point of view of the visual perception to be controlled (the distance between the cursor and the target) to the point of view of the proprioceptive perception that is used to control it (the effort exerted on joystick) we should see this shift reflected in a change in the variability of the parameters of control for the systems involved in controlling each of these variables. The real-time record of these changes would be a behavioral record of shifts in the point of view of the person’s awareness while performing the task. To echo Powers from four decades ago, this seems like an area of research that is worth pursuing since it could provide a “… scientific approach that might reveal some property of consciousness” (Powers, 1980, p. 242, italics his). In this case, the property of consciousness that could be revealed is its role in initiating the reorganization process.1

6.6  Summary Control is a capability that has to be learned. In PCT learning is called reorganization and it is assumed to be a biased random-walk process, similar to the process used by the E. coli bacteria to navigate to food sources. Research on reorganization is aimed at showing how this reorganization process works in the context of a hierarchical model of purposeful behavior. One example of research on reorganization is the Robertson and Glines experiment that involved monitoring the process of learning to control 1

The idea that conscious awareness guides reorganization is challenged by other anecdotal evidence that “after focusing on a problem for a period of time, the solution seems to ‘pop’ into awareness while [a person is] thinking of something else entirely – that is, while their consciousness is somewhere else” (Carey, 2020, pers. comm.). Clearly, more research is needed in order to get a clearer picture of the role of consciousness – awareness and volition – in learning (reorganization).

6.6  Summary

103

three different, nested variables. The results showed how each level of control was learned in order, from lowest to highest, each after a period of random reorganization. Another approach to the study of reorganization is the naturalistic observation of the development of skills in infant chimps and humans. Evidence from observational research carried out by the Plooijs suggests that both chimp and human infants learn skills in a predictable sequence that is consistent with the order of levels of control, from lowest to highest, in the PCT hierarchy of control systems. PCT also makes predictions about the way consciousness is involved in the reorganization process. We described one possible way to investigate the role of consciousness in reorganization. This involved monitoring the parameters of control in a multilevel control task while the person doing the controlling moves their conscious awareness to different points of view of the task. The goal is to see whether (and how) consciousness induces reorganization in the form of increased variability of the parameters of a control system that is being consciously monitored.

7

Social Control

The research we have discussed so far is aimed at understanding the purposeful behavior of individuals taken one at a time. But some of the more important purposeful behaviors we see are produced by groups of individuals controlling together in some way, a phenomenon that can be called social control. Familiar examples of social control include the flocking of birds, the hive building of bees, and the orchestral performances of humans. Like Tolstoy’s happy families, all social control is the same inasmuch as it always involves two or more living control systems interacting with each other. But there are several different ways in which these interactions can take place. There are differences in how the controlling done by each individual affects the controlling done by others in the group as well as how each individual’s controlling contributes to the results produced by the group as a whole. The different ways individual control systems interact with each other result in different kinds of social control phenomena. We will discuss three kinds of social control that have been studied from a Perceptual Control Theory (PCT) perspective; that is, from the perspective of viewing individuals in the group as purposeful (control) systems. These three types of social control can be described as cooperative control, conflictive control, and manipulative control.

7.1  Cooperative Control Cooperative control occurs when two or more living control systems act together to produce results that could not have been produced by a single system on its own. For example, you are doing cooperative control when you join with a friend to lift a heavy piece of furniture, such as a couch, that you could not lift on your own. In order to do this, both you and your friend must cooperate by coordinating your lifting efforts so that the couch gets lifted. Another example of cooperative control is the V-formation of flocking geese. Again, in order to produce this result, the geese in the flock 104

7.1  Cooperative Control

105

have to coordinate their flight paths so that they remain in their proper positions relative to each other. 7.1.1  Modeling Two-Person Cooperation Research on cooperative control is aimed at understanding how the individuals in a group are able to precisely coordinate their activities so as to produce a result – such as a lifted couch or a V-shaped flocking formation – that could only have been produced cooperatively. The main research methodology involved in the study of cooperative control is the use of modeling. Models of cooperative control view the individuals involved in cooperative interactions as control systems that are controlling one or more perceptual variables. The goal of modeling is to build models of the controlling done by the individuals in a group so that the cooperative control phenomenon under study emerges from their interaction. A nice example of this approach to understanding cooperative control is research on cooperation between pairs of individuals that was carried out by Tom Bourbon (1990). Bourbon developed a model that accounts for the cooperative behavior in a two-person compensatory tracking task that is analogous to the task of lifting a couch. The two participants control two separate cursors and are asked to “lift” them to the same height on the screen. The task is set up so that each participant’s output (handle movements in this case) affects their own cursor as well as that of the other participant. The two cursors are analogous to the two ends of the couch that is being lifted and the effect of each participant’s output on the other participant’s cursor is analogous to the effect of lifting one end of the couch on the output required to lift the other end. The model that accounts for the behavior in this task consists of two separate control systems, representing the two people in the tracking task who are doing the equivalent of lifting a couch. Each control system is trying to lift its end of the “couch” to the same height. To do this, both systems have to cooperate by coordinating their outputs appropriately to simultaneously lift both ends of the couch to the same height while compensating for the disturbance created by the effect of the other system’s outputs on their side of the couch. The systems can do this because each one is “automatically” varying its output to compensate for the disturbance created by the other so that both cursors (both ends of the “couch”) are lifted in concert. The mutual compensation of each system for the other’s output is what Bourbon calls a cooperative “dance” that is choreographed by the disturbance-resisting characteristic of the control systems.

106

Social Control 7.1.2  Modeling Group Cooperation

Other models of cooperative control deal with cooperation between more than two individuals, such as the cooperation between hundreds of flocking birds. A flock is a moving formation of a group of birds. It is a cooperatively produced result inasmuch as the birds that make up the flock must coordinate their flight patterns with each other in order to maintain the formation. One of the original models of flocking was developed by Reynolds (1987). In Reynolds model, the cooperation that results in flocking formations is achieved by having each bird (called a boid in the simulation) control three different perceptual variables: (1) distance from neighboring boids, (2) alignment with the heading of neighboring boids, and (3) movement toward the center of mass of neighboring boids. This model produces flocks that move in formations that resemble those seen in many different species of bird. As in Bourbon’s two-person model of cooperation, the coordination of outputs (flight patterns) seen in the models of flocking behavior is a result of the disturbance-resisting characteristic of the control systems in the boids. Since each boid is controlling a perception of itself relative to other boids, the disturbances in the flocking models are created by the movements of the boids relative to each other. These movements would “break up” the flock formations if they were not being continuously countered by the actions of the boids themselves. So, again, the cooperative “dance” that is the flocking behavior of boids is choreographed by the disturbance-resisting characteristic of the control systems that make up each member of the flock. Another model of cooperation between several individuals was built to explain the spatial arrangements that are formed by groups of people who get together in various types of temporary gatherings (McPhail, Powers, and Tucker, 1992). Examples of such arrangements are the line that forms to buy tickets to a concert or the semicircle that forms around a performer on the boardwalk. The model that explains these phenomena is called the CROWD program (Powers, 2005a, pp. 150–160). The model consists of a set of simulated people – the CROWD. Each person in the CROWD is made up of three control systems; one system controls a perception of the person’s proximity to neighboring people and objects, another system controls a perception of following a target person and a third system controls a perception of moving to a specific destination. By adjusting the parameters of the control systems in each person, it is possible to create scenarios where the people in the CROWD arrange themselves into different forms of temporary gatherings.

7.1  Cooperative Control

107

An example of the kind of scenario that can be created by the CROWD program is shown in Figure 7.1. In this scenario, a group of eight “acolytes” follow a “guru” through an array of obstacles to a point where the guru stops. Figure 7.1 shows the final state of this “guru” scenario. The guru is the gray dot inside the larger circle on the right. The white dots in a semicircle around the guru are the acolytes and the black dots are the array of obstacles. The figure also shows the paths that the guru (solid line) and three of the acolytes (dashed lines) took to get to the final state. The guru is made up of two control systems, one controlling a perception of maintaining a fixed distance from other people (the acolytes) and objects (obstacles) and the other controlling a perception of arriving at a specific destination. That destination is indicated by the circle that surrounds the guru. Each acolyte is also made up of two control systems. Like the guru, one of these systems is controlling a perception of maintaining a fixed distance from other people and objects and the other is controlling a perception of tracking a target person that, in this case, is the guru. The simulation starts with the guru and acolytes in random positions on the left side of the display. The guru and acolytes then make their way through the array of obstacles, with the guru leading the way. Once the guru arrives at the destination, the acolytes arrange themselves around him (or her), equally spaced from one another in a perfect semicircle. Like the formations formed by flocking boids, the semicircle formed by the acolytes is a cooperatively produced result that is choreographed by the disturbanceresisting characteristic of the control systems that make up each of the participants – guru and acolytes – in this scenario.

Figure 7.1  Tracks to the final state of the “guru” scenario of the CROWD simulation.

108

Social Control 7.1.3  Cooperation and Culture

Culture refers to the ways people in different groups carry out the same activities. For example, cooking is something that all people do, but it is done differently by people in different groups; American cuisine differs from Japanese cuisine which differs from French cuisine and so on. The way cooking is done by the people in a particular group is a characteristic of the group’s culture. These cultural characteristics can be considered cooperative phenomena since what we see are people “going along” with what others are doing rather than “doing their own thing.” One theory of how these different cultural characteristics develop is that it occurs via a process that is analogous to the evolution of different physical characteristics (phenotypes) in different species. What is needed to make this analogy work is a means of transmitting cultural traits from one generation to another in the same way that genes transmit physical traits from one generation to another. Dawkins (1989) suggested that if such a mechanism exists, it should be called a meme. Like a gene, the meme would be a carrier of information, but information about a cultural rather than a physical trait. And this information would be passed from one person to another, mainly by imitation. While attractive because of its derivation from evolutionary theory, the meme theory of cultural evolution suffers from lack of a clear description of the mechanism of memetic selection. However, there is some evidence that PCT may provide that mechanism. The evidence comes from research aimed at understanding the development of regional differences in language pronunciation. Differences in the way the same language is pronounced by different groups – such as the different way English is pronounced by Southerners and Northerners – can be considered a difference in a cultural trait of the groups. One study of such differences was done by Labov (1963) who obtained quantitative measures of pronunciation of the diphthongs /ai/ and /au/ from natives of two different regions of the island of Martha’s Vineyard, Up Islanders and Down Islanders. The measure of pronunciation was called the centralization index (CI). CI is a measure of the height of the tongue when pronouncing the diphthongs; the closer the tongue to a central position in the mouth when pronouncing the diphthongs, the greater the CI value. This difference in the centralization of the tongue corresponds to a difference in how the diphthongs sound. Labov found that there was a substantial difference between Up and Down Islanders in their average CI index for both /ai/ and /au/. For Up Islanders, the average CI index was 0.61 for /ai/ and 0.66 for /au/; for Down Islanders, the average CI index was 0.33 for /ai/ and 0.35 for /au/.

7.1  Cooperative Control

109

A simple PCT model was developed to account for the evolution of the different styles of pronunciation in the two groups of speakers. The model was based on the assumption that people control for pronouncing diphthongs in the same way that they are pronounced by the people with whom they interact. The model starts off with a population of people who all pronounce the same diphthongs slightly differently. The different pronunciations are different CI values that are randomly assigned to the different members of the population. This population is then divided into two subpopulations corresponding to Up Islanders and Down Islanders. The individuals in each subpopulation interact only with each other and not with members of the other subpopulation. Interactions between members of each subpopulation consisted of pairs of members controlling for imitating the pronunciation (CI value) of the other. They did this by varying their output (vocal production) so that their own pronunciation approaches that of the other party to the interaction. That is, each member was acting so as to move their own CI value toward that of the member with whom they were currently interacting. Therefore, after each interaction, each member of a subpopulation had changed their own pronunciation (their own CI value) to be more like that of a different member of the subpopulation. After many of these imitative interactions, the average diphthong pronunciation (CI value) in each subpopulation converged to a single value that remained stable over time. These results are shown in Figure 7.2. The figure shows the results of one simulation run of the PCT model of the evolution of pronunciation styles. The simulation starts with the average pronunciation styles (average CI values) of the two subpopulations – Up and Down Islanders – being equal. As time goes on, the averages diverge as random interactions within each subpopulation result in differences in which CIs are most frequently imitated. After a period of variation, the average CI value in each population stabilizes at some value. In this run, the average for the simulated Up and Down Islanders stabilizes at values that happen to be close to the values observed by Labov. Due to the random nature of the interactions, the average CI values for Up and Down Islanders stabilize at different values on different runs. But they do always stabilize, making this simple PCT model a plausible start at a model of the evolution of different cultural styles. The equivalent of a meme in this model is what has been called each member’s CI value. These CI values are actually the reference specifications sent to the lower level systems that are responsible for controlling the articulatory perceptions that are responsible for pronouncing the

110

Social Control

Figure 7.2  Convergence to different styles of pronunciation for Up and Down Islanders on Martha’s Vineyard through control of imitation.

diphthong (CI value) the way another person pronounces it. Of course, this simple model doesn’t actually do this; it simply produces a CI value that approximates the reference for that value. Eventually, however, this level of complexity should be added to the model. But even this simple version of the model provides some evidence that the meme that carries the specifications for different cultural styles – such as the different styles of pronouncing diphthongs – is a reference signal in the brain of each member of the culture that specifies what the diphthong should sound like. And a likely mechanism for the development of these memes and passing them from one generation to another is the existence of an innate tendency to control for imitating the pronunciation of other members of the group. 7.1.4  Agreeing to Cooperate The models of cooperative control described above show how the PCT model of individuals can be used to study the cooperative behavior of groups of individuals – behavior that produces results that could not have been produced by any individual acting alone. However, all these models leave out one crucial step in the development of groups capable

7.1  Cooperative Control

111

of producing cooperative results: the initial agreement about what each individual in the group should control. This agreement is either made explicitly, via communication between the parties involved, or implicitly, via evolutionary or cultural selection. The cooperation that results in a lifted couch, for example, is a consequence of an explicit initial agreement based on communication between the parties involved. The parties have agreed to lift the couch together, each lifting from opposite ends of the couch and timing their lifts to be simultaneous so that the load lifted by each party is about the same. Bourbon’s model of the control systems that lift the couch assumes that these agreements have already been made (and, indeed, they were implicitly made in the two-person tracking task when the two participants agreed to follow the instructions given by the experimenter). The cooperation that results in the flocking behavior of birds must also have been based on an “initial agreement” to flock in formation. But in this case this agreement was arrived at by evolutionary selection. The evolutionary origins of the agreement to flock is suggested by the fact that flocking is known to provide significant survival advantages to individual birds in the flock, not the least being a reduced demand for power and energy expenditure during flight (Weimerskirch et al., 2001). So birds born with an inclination to control variables like those controlled in Reynolds’ flocking simulation – variables that, when controlled, result in the birds flying together in aerodynamically advantageous flocking formations – would have a selective advantage over those born without such an inclination. The result is an evolutionarily selected cooperative control process – flocking. Evolution may also be involved in a group’s “initial agreement” to develop a cultural style, such as a style of speaking. The agreement would be to imitate the behavior of others in the group. This “agreement” seems to be present very early in human development, suggesting that it may be an inherited capability of the human control hierarchy. Other evidence of this is the discovery of mirror neurons (Keysers, 2010). These are neurons that fire both when an animal sees some behavior and when it performs the same behavior itself. Since, from a PCT perspective, behavior is a controlled perceptual signal, mirror neurons are likely to be the same perceptual signals that are produced when an animal passively observes a behavior and when it produces the same behavior itself. But whether the inclination to imitate is inherited or learned, it can serve as the initial “agreement” that is the basis for the cooperative development of the cultural styles that characterize different groups of people.

112

Social Control

The cooperation that results in the spatial arrangements seen in temporary gatherings – and simulated by the CROWD program – are also likely to be a consequence of an “initial agreement” based on something similar to evolutionary selection. But the selection is more likely based on cultural rather than biological survival. The spatial arrangements in temporary gatherings result from the control of perceptual variables that are relevant to a person’s interactions with the other people who make up a person’s immediate cultural environment; variables such as maintaining an acceptable distance from others and following people designated as leaders. People probably learn to control these variables in order to maintain control of their interactions with parents and peers. And the side effects that result from controlling these variables are the spatial arrangements in temporary gatherings such as the ones seen in Figure 7.1. 7.1.5  The Social Contract When an organism agrees to cooperate, it is agreeing to let itself be controlled in order to produce a result that it could not produce on its own. For example, one of the two people lifting the couch had to agree to be controlled by the other person’s “lift” signal so that upward force could be applied simultaneously to each end of the couch; each flocking bird had to “agree” to let its direction of flight be controlled by the birds around it in order to get the survival advantages of being in the flock. Put another way, cooperation involves agreeing to give up some autonomy in order to achieve some greater benefit for oneself and the group as a whole. This is what the philosopher Jean-Jacques Rousseau referred to as a social contract (Gourevitch, 2018). It is the formal or tacit agreement to give up some individual control in order to get more control for oneself and one’s society. The agreement to lift simultaneously from each end of the couch on a signal from one of the lifters is an example of a formal agreement. It is formal because the agreement is produced “on purpose” and documented in a verbal or written statement. An example of a tacit social contract is the “agreement” made by each bird in the flock to maintain a fixed distance from nearby birds. It is tacit because it was not produced on purpose or documented in any way. Because cooperation requires giving up some control by allowing oneself to be controlled, there is considerable risk involved in agreeing to form a social contract. This is particularly true when making formal social contracts. The risk is that the parties to such a contract can cheat by not fulfilling their part of the agreement, thus getting the benefit of the cooperatively produced result without contributing their share. An example is people who avoid

7.1  Cooperative Control

113

paying taxes while enjoying the benefits of public services. The fact that such cheating occurs is one reason why it is so difficult to produce formal agreements to cooperate. Since the models of cooperative behavior that were described above assumed that the agreement to cooperate had already been made, these models could be seen as testing hypotheses about what these agreements were – what variables the parties to the agreement had agreed to control and not control – rather than how these agreements were arrived at. Therefore, these models fail to explain one of the most important aspects of cooperative behavior – the process of developing a cooperative agreement. Since the development of formal agreements to cooperate – social contracts – is a purposeful process, the question of how this process is carried out seems to be one that should be addressed by research on purpose – research based on PCT. Such research could be aimed at determining how the types of perceptual variables controlled by those developing the agreement affect how easily agreement is reached. Such research could also examine how the types of variables specified as those to be controlled by parties to the agreement affect how well the agreement is kept. Studying the process of forming social contracts from a PCT perspective could lead to useful insights about how groups of people can form “more perfect unions.” 7.1.6  Conflictive Control Conflictive control is just the opposite of cooperative control. While cooperative control involves individuals working together to achieve a result that each individual could not produce on their own, conflictive control involves individuals or groups working against each other to achieve a different result for each individual or group. From the control theory perspective, conflict occurs when two or more control systems are trying to bring the same perceptual variable to different reference states. When these control systems are in the same individual, the conflict is intrapersonal; when the control systems are in different individuals the conflict is interpersonal. We talked about intrapersonal conflict in Chapter 6 when we discussed the role of reorganization in psychotherapy. The conflict that occurs in the conflictive form of social control is interpersonal; the conflicted control systems are in different individuals. One example of interpersonal conflictive control is arm wrestling, where two individuals try to pin their clasped hands to opposite sides of the table. The variable in conflict is the position of the clasped hands and the different reference states of that variable are the opposite sides of the table where each individual wants to pin their opponent’s hand. Another all too familiar example of conflict is war, which often occurs when two different

114

Social Control

groups – tribes, nations, etc. – try to occupy the same piece of land. In this case, the variable in conflict is who occupies the land and the different reference states are the different ideas about who it should be. Conflictive control like this probably shouldn’t be called “control” since some – and often all – parties to the conflict are not getting what they want; they are not able to control the variable that is in conflict. In arm wrestling, when the clasped hands are not pinned to either side of the table neither wrestler is in control; when the hands are finally pinned to one side or the other, the winning wrestler is in control and the loser is not. The same applies to warring groups; the group that currently occupies the land is in control until it is displaced by the other group. But while some or all of the parties to a conflict are not in control, the conflict itself is a result of the controlling done by the parties to the conflict. So, paradoxically, controlling can lead to the loss of control (Marken and Carey, 2015). The behavior seen in conflictive interactions between control systems differs depending on the relative strengths of the systems involved in the conflict. The differences can be described in terms of the behavior of the variable that is the object of the conflict. For example, in an arm wrestling conflict, the object of the conflict is the position of the clasped hands. The behavior of this variable depends on the relative strengths of the arm wrestlers. The strength of a control system is measured in terms of gain and maximum output. Gain is the amount of output the system produces per unit error and maximum output is the maximum amount of output the system is physically capable of producing. 7.1.7  Conflict between Control Systems of Equal Strength When the systems involved in a conflict are of equal strength, in terms of both gain and maximum output, the controlled variable remains in a virtual reference state (Powers, 1973b, p. 255; McClelland, 2004). For example, when arm wrestlers are of equal strength, the position of their clasped hands oscillates in a narrow band perpendicular to the table. The average position of this oscillation is the virtual reference state of the clasped hands and the position of the clasped hands is called a virtual controlled variable. Like an actual controlled variable, a virtual controlled variable is kept in a reference state protected from the effect of disturbances. If, for example, you came by and pushed the clasped hands toward one of the competitors that push would be resisted. (I would recommend not trying this yourself unless you are good friends with the competitors and tell them in advance what you plan to do.) So the clasped hands appear to be controlled because they are being kept

7.1  Cooperative Control

115

in a reference state. But this reference state is virtual, not actual, because it is not the reference state specified by either competitor. When a variable is being kept in a virtual reference state by a conflict, none of the parties to the conflict are getting what they want; none have the variable under control. The existence of virtual reference states means that variables can appear to be controlled variables when they are not. They pass the Test for the Controlled Variable; disturbances to these variables will be resisted. But this is only true for disturbances that push the variable out of the narrow band in which they oscillate. This band is called the dead zone of the conflict because disturbances that occur within this zone will not be resisted. This fact provides a good basis for distinguishing virtual from actual reference states of controlled variables. Controlled variables that are being maintained in actual reference states are protected from disturbances of all sizes while those being maintained in virtual reference states are protected only from large disturbances that push the variable out of the dead zone. You can tell that the clasped hands of the arm wrestlers are being kept in a virtual rather than an actual reference state because strong pushes are resisted while small ones are completely effective. Of course, this approach to distinguishing virtual from actual reference states is unnecessary if you can see that there is a conflict going on. This is the case in the study of the robbing and dodging behavior of rats and other organisms. It has been found that a rat holding food, known as the dodger, maintains a fairly constant distance between itself and another rat trying to get the food, known as the robber (Bell, 2014). The distance between the dodger and robber appears to be the reference state of a controlled variable. But we know that the distance between the robber and dodger is being controlled by two different control systems – the robber and the dodger – who want that variable in different states; the robber wants that distance to be zero and the dodger wants it to be considerably greater than zero. Thus, the constant distance between the robber and dodger is likely to be the virtual reference state of a variable that is in conflict. And, indeed, a model which views the robber and dodger as control systems with different reference specifications for the distance between them accounts for the observed distance quite well (Bell and Pellis, 2011). 7.1.8  Conflict between Control Systems of Differing Strength Control systems that come into conflict are rarely of equal strength, either in terms of gain or maximum output. And even if the control systems are of nearly equal strength at the start of a conflict, their relative strength

116

Social Control

will vary over time so that one may eventually be able to win the conflict. Relative strength varies over time because control systems differ in endurance – the ability to maintain their strength over time. Once one control system becomes sufficiently stronger than the other, it can overcome the disturbance created by the now weaker system’s output and force the controlled variable to its desired reference state. This is why there is almost always a winner and loser in an arm wrestling conflict. The wrestlers may start at equal strength but eventually one wrestler’s endurance will run out and the wrestler with better endurance will overcome the now weaker one and force the clasped hands down on the opponent’s side of the table. The “winner” of a conflict has successfully controlled the variable that was the subject of the conflict. So winning a conflict seems to be a case where being in a conflict has not impaired the ability to control, at least for the stronger control system. But this is only true for conflicts that have agreed upon endpoints, such as arm wrestling matches and football games. When the endpoint is reached – when one of the contestants wins – the parties to the conflict stop trying to control the variable that had been in contention; the arm wrestlers stop controlling the position of the clasped hands and the football teams stop controlling the location of the football. In real social conflicts, such as wars, there are no agreed upon endpoints so the variable or variables in conflict can remain in contention even after there is a winner. While the winner is enjoying victory, the loser is regaining strength so that “the loser now will be later to win” (Dylan, 1963b); the winner’s control is only temporary. Interpersonal conflicts can produce positive as well as negative results. Athletic contests are interpersonal conflicts that are enjoyed by both spectators and players. Scientific debates are interpersonal conflicts that can lead to new discoveries and better theories. Legal arguments are interpersonal conflicts that can be a relatively peaceful alternative to violent confrontation. But PCT shows us that, in general, conflicts result in some loss of control by the parties involved. This is true whether the parties to a conflict are of equal or unequal strength. One goal for research on conflict would be to see if there are ways to resolve such conflicts so that all parties remain in control. PCT suggests a method for resolving a conflict – a process called “going up a level” – that was discussed in Chapter 6 in the section on reorganization in psychotherapy. In that discussion, it was noted that a conflict between control systems within the same person – an intrapersonal conflict – can be solved when the person becomes aware of the conflict from a higher level point of view. Solving a conflict between control systems when these systems are in different people – an interpersonal conflict – is far more challenging.

7.2  Manipulative Control (Control of Behavior)

117

All parties to such a conflict would have to become aware of that conflict from the same higher level point of view at the same time. Research on solving interpersonal conflict could be aimed at figuring out how to get this to happen. Some preliminary efforts in this direction have been made using an innovative approach to theater where audience members are invited to intervene in conflict scenarios, demonstrating alternative points of view to all parties to the conflict at the same time (Scholte, 2019).

7.2  Manipulative Control (Control of Behavior) Manipulative control is a form of social control that occurs when control systems try to control each other’s behavior. It is of particular interest in psychology when the control systems involved are human beings. The idea that human behavior can be controlled is somewhat controversial; some think it can be controlled (e.g., Skinner, 1971) and others think that it can’t (e.g., Koestler, 1990b). PCT suggests that behavior can be controlled, but only under some circumstances. It also suggests that it is generally a bad idea to try to control a person’s behavior without their consent since doing so is likely to result in conflict. Both of these predictions about the possibility and consequences of controlling human behavior are worthy of further research, so I will describe how these predictions are derived from PCT. 7.2.1  Manipulating Behavior PCT shows that it is possible to manipulate (or control) behavior by taking advantage of the disturbance-resisting characteristic of control systems. The disturbance-resistance approach to manipulating behavior can be demonstrated using the rubber-band demo discussed in Chapter 1 (Figure 1.3). The rubber band demo involves two people, dubbed E and P (experimenter and participant, respectively), pulling on opposite loops of two rubber bands that are knotted together. P is to keep the knot over a target mark by pulling on his loop of the rubber bands. The pulling can be done using a finger or pen (as in Chapter 1) but let’s assume that P is using a finger. E can now demonstrate control of P’s behavior by disturbing the variable P is controlling – the distance from the knot to the target dot. In order to keep the knot on target (in PCT terms, in order to keep the controlled variable in the reference state of zero distance from the target), P’s finger movements must mirror E’s. So E can control the behavior of P’s finger – moving it to some prespecified location or having it trace out a

118

Social Control

particular pattern – by pulling on her end of the rubber bands appropriately. E is able to do this “control by manipulation” because, as Powers puts it: P cannot control both his finger and the knot: they are connected. Therefore, if P wants to control the knot, E can control P’s finger, as long as the result does not inconvenience P (as by running the finger into a hot soldering iron). (Powers, 1973b, p. 243)

In this brief quote, Powers implicitly describes the conditions under which control of behavior will succeed and when it will fail. It succeeds when the controller (E in this case) knows what variable the controlee (P in this case) is controlling and the outputs P must use to control it. In the rubber band demo, E knows that P is controlling the distance between the knot and the target dot and that the output P must use to control that variable is pulling on his end of the rubber bands. As long as P wants to control the distance between the knot and the target dot, E can control P’s rubber band pulls (which is the only aspect of P’s behavior that E can control). E’s attempts to control P will fail, however, when, for whatever reason, P stops controlling the variable E was disturbing as the means of manipulating P’s behavior. The Powers’ quote also hints at why attempts to control behavior can create problems for the controlee. In the rubber band demo, these problems are created when E tries to get P to do something that conflicts with one of the P’s other goals, which was avoiding pain. The problem this creates for P is conflict; the conflict is between keeping the knot on the target dot and not getting burned. P can’t do both at the same time. In this case, the conflict is probably easily resolved in favor of not getting burned since keeping the knot on target is not likely to be particularly important to P. But in real life, conflicts created by attempts to control behavior can be very debilitating for the controlee, particularly when the carrot–stick approach is used to control behavior. 7.2.2  Carrots and Sticks and Conflicts The carrot–stick approach to behavior control involves making rewards (carrots) or punishments (sticks) contingent on particular behaviors. An example of using a “stick” to control behavior was the shock avoidance study described in Chapter 2. In that study, the rat could avoid shocks only by pressing the lever a sufficient number of times in a fixed interval. A controller could then control the rat’s rate of lever pressing by varying the duration of that interval – the shorter the interval the higher the rate of pressing. An example of using a “carrot” to control behavior is the operant

7.2  Manipulative Control (Control of Behavior)

119

conditioning experiment, described in Chapter 4, where the controlee is given a food pellet (reinforcement), only after producing a sufficient number of lever presses. Again, the controller can control the rat’s rate of lever pressing by varying the number of lever presses required to get the food pellet – the more presses required, the higher the rate of level pressing. Both the carrot and stick approaches to controlling behavior have been used successfully to control the behavior of all kinds of organisms, including humans (e.g., Liberman, 1971; Lovaas, 1987) The problem with using rewards and punishments to control behavior is that they can put the controlee in a conflict that is very difficult to solve. For example, if a parent tries to control a child’s behavior by making avoidance of the “stick” of punishment contingent on the child performing unpleasant tasks, the child is being placed in a difficult conflict. If the child avoids punishment by performing the tasks, there will be an error in the system inside the child that is controlling for not doing the tasks; if the child doesn’t perform the unwanted tasks, it will be punished, resulting in error in the system controlling for not being punished. If the child doesn’t resolve this conflict, it will suffer from stress and possibly other health problems that result from the persistent error that exists in the control systems controlling for avoidance of punishment and for not doing the unpleasant tasks. And if the child does resolve the conflict, it will not necessarily be in a way that is satisfactory to the controller. This analysis of manipulative control suggests several possible lines of research on purpose. One possibility is to see how the ability to control behavior using rewards or punishments depends on how well those rewards or punishments can be controlled by the controlee. The PCT analysis of manipulative control shows that people are most easily controlled when they have good control of the variables the controller is using to control them. For example, the better P can control the distance between the knot and dot in the rubber band demo, the more effectively E can control P’s behavior by disturbing that variable. The results of this research could inform people about how to protect themselves from manipulation by would-be controllers, such as propagandists and demagogues. It would also be useful to see if the results of controlling behavior differ depending on whether this controlling is done with or without the concurrence of the controlee. In this discussion of manipulative control, we have assumed that the controller exerts control arbitrarily – without regard to what the controlee actually wants and without getting the controlee’s agreement to be controlled. It’s likely that the controlee’s behavior can be controlled more precisely – and without the creation of conflict – when the controlee agrees

120

Social Control

to be controlled. Indeed, as discussed in the section on the social contract, it’s this kind of agreement that is a prerequisite to doing cooperative control. 7.2.3  It Takes a Controller Missing from most discussions of manipulative control – particularly from discussions by those who believe that such control is not only possible but essential (Skinner, 1971) – is the fact that, in order to control behavior, controllers must be control systems themselves. Controllers must (1) be able to perceive another person’s behavior, (2) have a reference specification for what is the “right” state for that behavior, and (3) be able to act in such a way as to bring the controlee’s behavior to the reference state desired by the controller and keep it there, protected from disturbances (which occur when the controlee does something other than what the controller wants done). Although this kind of behavior is generally considered problematic – no one likes people who are controlling – the fact is that we are all controllers and, therefore, we are often in situations where we want to control other people. We want to control our children so that they behave decently, we want to control our politicians so that they support the right policies, and so on. So it seems that the study of behavior controllers – the people (like all of us) who control other people – would be a worthwhile subject of research by those interested in social control. What we would want to find out is what perceptual aspects of other people’s behavior do controllers control; what are the reference states of these perceptual variables and what are the means used to get these variables into those reference states. The aim of such research would be to find ways for people to interact with each other in ways that reduce the chances of conflict. That is, the aim would be to find out how humans who are naturally inclined to control perceptual aspects of their environment – including the behavior of other people in that environment – can live together in something approximating peace and harmony.

7.3  Summary The social interactions between control systems can be categorized into three types: cooperative, conflictive, and manipulative. Cooperative interactions involve the production of a result that can only be produced by two or more control systems working together. We described models of several different examples of cooperative interaction: lifting a couch, birds flocking, patterns formed in temporary gatherings, and “cultural” differences in pronunciation. These cooperative results require a formal or tacit social contract, which is an agreement about what members of the group will control in order to produce

7.3  Summary

121

those results. Conflictive interactions occur when two or more control systems want the same variable or variables in different reference states. The result of such conflicts can be that the commonly controlled variable appears to be under control. It is important to be able to distinguish these virtual controlled variables from variables that are actually being controlled by members of a collective. Manipulative interactions occur when one control system tries to control the behavior of another. These interactions will result in inter- and intrapersonal conflicts unless the controller and controlee can agree to the terms of their relationship, in which case the manipulation becomes a form of cooperation.

8

Back to the Future (of PCT Research)

The research methods and the examples of research using these methods that are described in this book are meant to show you what a research program aimed at testing the Perceptual Control Theory (PCT) model of behavior might look like. Powers described his own vision of what such a program might look like in an essay on a control theory approach to doing research on human development (Powers, 1979c) – an essay written shortly after the publication of his monograph introducing his control theory model of behavior that has come to be called PCT. It’s a vision of a research program that differed considerably from the way behavioral research was done at that time and the way it is still being done. Indeed, Powers’ vision of a PCT-based research program differed so much from the way research was being done that only a few researchers were willing to try it out. At least a generation has passed since Powers described his vision of a PCT research program and it is my hope that, like young Marty McFly’s rock-and-roll riff that baffled the audience at his Mom’s prom in the movie Back to the Future, the current generation of researchers may be ready to not only give it a try but to like it. This chapter will take us back to Powers’ vision of the PCT-based research program of the future, not only because we are now well into that future but also because doing so provides a “big picture” view of what is involved in doing research on purpose.

8.1  A Control Theory Model for Psychological Research Powers’ vision of a PCT-based research program consisted of two main stages: data gathering and data classification. 8.1.1  Data Gathering The principal kind of data needed for a PCT-based research program concerns what variables organisms control. Powers’ vision of data gathering 122

8.1  A Control Theory Model for Psychological Research

123

involved the creation of “… catalogs of controlled variables which have passed every application of the test for the controlled variable, conducted with as much rigor as possible” (Powers, 1979c, p. 216). The controlled variables in these catalogs could come from laboratory studies of behavior, such as those described in Chapter 4, where the test for the controlled variable (TCV) can be done with a high degree of rigor. But they could also come from studies of behavior in more naturalistic situations, such as the Plooijs studies of the behavior of infant chimps in Gombe National Park, where a less rigorous, but still informative, version of the test is possible (Plooij, 1987). These data would be gathered to test two basic tenets of PCT: (1) that behavior is the control of perception and (2) that controlled perceptions are organized in a hierarchy with simpler, lower level perceptions controlled as the means of controlling more complex, higher level perceptions. The first tenet is tested by demonstrating the existence of controlled variables using the TCV; if we were unable to demonstrate the existence of such variables using the TCV “… we [could] save ourselves a lot of trouble by abandoning the project there” (Powers, 1979c, p. 216). But as has been demonstrated throughout this book, researchers have identified many of the variables organisms control when they are carrying out various behaviors, providing strong evidence that behavior is, indeed, organized around the control of perceptual variables. The second tenet is tested by seeing whether the controlled variables in the catalogs are consistent with the hierarchical organization of controlled perceptions proposed by PCT. In particular, the goal is to see whether the types of controlled variables in the catalogs are consistent with the types of perceptual variables that are proposed as being controlled at the different levels of the PCT hierarchy. In order to do this testing properly, the controlled variables entered into the catalogs should be selected “… without regard to any proposed hierarchy or any other preconceived notions” (Powers, 1979c). The point of this caveat is to make sure that researchers treat the proposed hierarchy as a hypothesis rather than holy writ. While Powers and others have presented behavioral and neurophysiological evidence for the types of variables controlled at each level of the hierarchy of control systems, the proposed hierarchical model must still be considered a hypothesis. An example of what a catalog of controlled variables might look like is shown in Table 8.1. This table shows just a small portion of the data that were gathered in an online computer exercise from an international array of students of PCT. The students were asked to describe familiar examples of behavior that came randomly to mind. They were then asked to describe

Table 8.1 A portion of a possible catalog of controlled variables Behavior

Variable

Reference state

Means

Disturbances

Sweetening tea

Sweetness of tea

Not too sweet

Add sugar to tea

Form of sugar (cube, granulated)

Adjusting brightness of laptop display

Luminance level

200 Nits

Press brightness adjustment keys

Environment luminance level

Flossing teeth

Amount of food between teeth

No food

Pull floss

Space between teeth, amount of food between teeth

Hammering nail into plank

Nail head height above surface of plank

Flush with plank

Hit nail with hammer

Humidity, outdoor temperature, hardness of plank

Rolling egg into nest

Pressure on back of bill

Pressure centered

Pulling back on egg

Gravity vector

Opening car door

Angle of door

80 degrees

Grasp, pull

Weight of door, angle of car

Sipping tea

Position of cup

Cup at lips

Lift, tip cup

Amount of tea in cup

Intercepting moving object

Derivative of optical angle Zero

Movement relative to the object

Trajectory of object

Adjusting rear view mirror

Displacement of rear window image

Zero displacement

Grasp, twist

Tightness of hinge, height in seat

Typing “Hello”

Sequence of letters

“Hello”

Tap keys

Resistance of keys, typos

Seeking employment

Employment status

Employed

Read want ads, go to employment office

No ads, office closed

8.1 A Control Theory Model for Psychological Research

125

their guess about one of the variables around which the behavior might be organized (the controlled variable), the reference state of that variable, the actions required to keep the variable in the reference state, and possible disturbances that would have to be resisted in order to bring the variable to the reference state and maintain it there. The students were to enter these data into a row of a shared online spreadsheet with the information about the behavior placed in the appropriate column: the description of the behavior in the column labeled “Behavior”; the guess about the controlled variable in the column labeled “Variable”; the description of the reference state of the controlled variable in the column labeled “Reference State”; the actions required to keep the controlled variable in the reference state in the column labeled “Means”; and possible disturbances to the controlled variable in the column labeled “Disturbances.” Table 8.1 provides a nice picture of what the results of the data gathering phase of a control theory research program might look like. However, not all the entries for the controlled variable were entered after having “… passed every application of the TCV, conducted with as much rigor as possible.” The controlled variable associated with only one behavior – intercepting a moving object – is a result of such rigorous testing. The data for the other behaviors were entered based on reasonable guesses about the variables that were being controlled and how it was carried out. Nevertheless, the creation of a catalog of controlled variables like this is useful for more than just showing what such a catalog might look like. The creation of Table 8.1 allowed the would-be control theory researchers who participated in the online data gathering exercise to accumulate experience with identifying the variables involved in control behavior, which Powers thought should be an essential component of the initial effort involved in gathering data for a research program based on PCT. Creation of the table also provides researchers with the experience of going from a description of behavior from the point of view of an observer (the description in the “Behavior” column of Table 8.1) to one from the point of view of the behaving system itself (the description of the controlled variable in the “Variable” column of Table 8.1). And the descriptions of controlled variables that were not determined by rigorous testing can, nevertheless, serve as the initial hypotheses in more rigorous tests to determine the variable(s) controlled when carrying out each behavior. After such tests are performed, the definitions of these variables will become much more precise. If possible the definitions should be mathematically precise – as precise as the rigorously determined definition of the variable controlled when intercepting a moving object as the derivative of an optical angle.

126

Back to the Future (of PCT Research) 8.1.2  Data Classification

The process of gathering data for a catalog of rigorously tested controlled variables will be an ongoing effort. However, once the catalog contains a reasonably large number of controlled variables, the next phase of a PCTbased research program can begin. This is the phase where an effort is made to see if the different controlled variables in the catalog can be grouped into classes or types and, if so, whether these types correspond to the types that are assumed to be controlled at the different levels of the PCT hierarchy. In order to do this properly, the classification process should be based not only on a large number of controlled variables but also on controlled variables that are involved in a wide range of different behaviors – from simple behaviors like opening the oven to more complex behaviors like baking the perfect sourdough bread. The most straightforward approach to grouping controlled variables into classes would be to do this in terms of reaction time; how quickly the organism reacts to an abrupt disturbance to the controlled variable. As noted in Chapter 5, there is evidence that disturbances to lower level perceptions are corrected more quickly than disturbances to higher level perceptions. There is also evidence that the time it takes to react to a disturbance is about the same for controlled perceptions that seem to be of the same type. For example, the time it takes to react to a disturbance to a sequence of tones is about the same as the time it takes to react to a disturbance to a sequence of shapes (Marken, 2002b). This implies that the times to react to disturbances for the variables in the catalog of controlled variables should fall into discrete clusters rather than vary continuously. Variables that fall into the same reaction time cluster could then be considered to be of the same class or type of perceptual variable. Another approach to grouping controlled variables into classes would be in terms of the age range at which each variable became controllable. Humans and chimps are known to develop the ability to control different kinds of variables at different ages (Plooij, 1987). So a program of research could be carried out where children of different ages are tested on their ability to control the different variables listed in the catalog of controlled variables. The tests would be similar to those done by Piaget and his associates (Piaget and Cook, 1952). Piaget created tasks to test whether children of certain ages possessed certain mental capabilities, such as the ability to correct errors in logical propositions. In the PCT version of these tasks, these mental capacities are equivalent to the ability to control complex perceptual variables, such as the logical status of a syllogism.

8.1  A Control Theory Model for Psychological Research

127

Still another possible approach to classifying controlled variables would be to treat the descriptions of controlled variables as “stimuli” to be judged using standard psychophysical scaling methods (Baird, 1978). The basic procedure would be to have people judge the similarity of all these “stimuli” to each other. For example, people would be asked to give a numerical estimate of the similarity of the sweetness of tea to the luminance level of a laptop display, the pressure at the back of the bill to a letter sequence, and so on. These similarity estimates could then be analyzed using multidimensional scaling techniques, such as cluster analysis (Kaufman and Rousseeuw, 2005). The result of the scaling analysis is a “picture” of the relative position of the controlled variables in a multidimensional psychological space. The goal is to see whether the controlled variables in this space form clusters and, if so, whether the variables in each cluster correspond to the types of controlled variables posited by the hierarchical PCT model of control. Of course, any clusters produced by the multidimensional scaling analysis are based on subjective judgments of similarity between the different controlled variable “stimuli.” So any clusters found by such an analysis are not an objective test of the hierarchical control model. However, such results could provide the basis for the development of more objective, experimental methods for classifying controlled variables into types. Another goal of the data classification phase of a control theory-based research program is to determine whether there is, indeed, a hierarchical relationship between the different types of controlled variables in the catalogs. A model for the kind of research needed to test for hierarchical relationships between controlled variables are the “portable demonstrator” tests described by Powers, Clark, and McFarland (1960b). These demonstrations are set up so that it is necessary to control one variable as the means of controlling another. The nested hierarchical relationship between the controlled variables is demonstrated by the fact that the response to a disturbance to the lower controlled variable is much faster than the response to a disturbance to the higher level variable. A more precise experimental implementation of such a control task was described in Chapter 6 as a way of investigating consciousness. An experiment could be set up so that different continuously varying disturbances are applied simultaneously to the lower and higher level controlled variables. The responses to these disturbances can be continuously monitored. What should be seen is the behavior of two systems simultaneously resisting disturbances to the variables they are controlling. The lower level system would be continuously acting to resist disturbances to the variable it is controlling while the higher level system is acting to resist disturbances to the variable it is controlling. Since the

128

Back to the Future (of PCT Research)

higher level system resists disturbances by varying the reference for the perception controlled by the lower level system, the disturbance resistance of the higher level system should be seen in the varying reference state of the variable controlled by the lower level system. That is, it should be possible to see two hierarchically nested levels of control operating simultaneously.

8.2  The Future of the Study of Living Control Systems The research program described in this chapter was Powers’ vision of what psychological research would look like in the future when it was based on an understanding of organisms as living control systems. Though we are now living in that future, it is not yet the one Powers envisioned. By and large, psychological research continues to be done the same way it has always been done: manipulating independent (typically environmental) variables to determine their effect on dependent (behavioral) variables. Indeed, the prevalent attitude among behavioral researchers seems to be that this is the only way to study behavior scientifically, which may be why many of those who saw merit in Powers’ theory failed to see that it implied a completely new approach to the study of behavior – one that was aimed at the discovery of the purposes rather than the causes of behavior (e.g., Carver and Scheier, 1981). Powers’ theory was seen as “just another theory of behavior” and, therefore, one that could be tested using the familiar methods of experimental psychology. In order to use these methods to test PCT researchers have had to ignore the central question about the behavior of living control systems, which is: “what are the variables organisms control?” (Marken, 2020). PCT shows that once this question is answered, it is possible to predict how various different independent variables will affect the organism’s “behavior” (the dependent variable). Independent variables are variables in the environment that can affect the states of controlled variables; dependent variables are the means used to keep these controlled variables in their reference states. Once you know a variable the organism is controlling – a controlled variable – you know how the organism will behave in response to all possible disturbances to that variable. The relationship between disturbances, d(t), and the responses to those disturbances, qo(t), is given by Eq. (1.3) in Chapter 1: qo(t) ≈ 1/F·d(t). So nothing is lost by moving from doing conventional independent–dependent variable research to doing research on purpose (using the TCV); researchers will still learn how independent variables are related to dependent variables. But something very important is gained:

8.2  The Future of the Study of Living Control Systems

129

knowledge of the variables organisms control which explain the existence of these relationships. The vehicle that can take psychological research back to the future Powers envisioned is not a DeLorean but, rather, the ability to see the behavior of organisms as the control of perception (Powers, 1973b, 2005b). Researchers who are able to look at behavior this way will almost necessarily be inclined to ask the right questions about behavior – the questions that should be asked in a program of research aimed at understanding the behavior of living control systems. Rather than asking what is causing an organism to perform a particular behavior, they will ask what perceptual variables are being controlled when the organism is performing the behavior. Rather than asking what causes the fielder to run to the ball, they will ask what variables the fielder is controlling by running to the ball; rather than asking what causes the beaver to build a dam, they will ask what variables the beaver is controlling by building a dam.1 That is, they will take the first step in the research program envisioned by Powers: gathering data about the types of perceptual variables organisms control. And, of course, they will have to use some version of the TCV, as described in Chapter 2, to answer these questions. The ultimate benefit that comes from doing research on purpose is that the results of such research can lead to the development of control theory models that show how the behavior of living organisms actually works. The identification of controlled variables is a necessary first step in the study of living control systems. But once you have identified some of the variables around which a particular behavior is organized, there are many other interesting things to learn about the systems controlling those variables. For example, you might want to learn how these systems developed. This involves the study of reorganization, as discussed in Chapter 6 on learning. Or you might want to investigate how the controlling done by these systems relates to the experience of emotion. A PCT-based theory of emotion is described in Chapter 17 of Powers (2005b). It views emotion as a cognitive interpretation of the physiological effects of persistent error in a control system. You could also investigate how consciousness is involved in the controlling done by living control systems. And, of course, once a sufficient number of controlled variables have been identified and placed in a database like that shown in Table 8.1, you could start working on developing ways to classify these variables and complete Powers’ vision of an objective test of his hierarchical control model of purposeful behavior. 1

Apparently, it is the sound of rushing water. Very typical PCT-like research shows that the beavers build dams to bring the amplitude of rushing water to zero by stopping its flow (Richard, 1983).

130

Back to the Future (of PCT Research)

8.3  Summary This chapter gives a general outline of Powers’ vision of a research program aimed at testing his PCT model of purposeful behavior. The program involves gathering data on the variables organisms control when they are carrying out various behaviors and classifying those variables into types. The goal of the program is to see if the variables organisms actually control correspond to the types of variables hypothesized by the PCT model and the hierarchical relationships between them. Once it is validated by testing, the PCT model can be used as a basis for understanding behavior in a way that opens up the possibility of developing better ways of dealing with the living control systems that are often the ones most prominent in our lives – other people.

Bibliography

Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgment. In H. Guetzkow (Ed.), Groups, leadership and men (pp. 295–303). Pittsburgh: Carnegie Press. Baird, J. C. (1978). Fundamentals of scaling and psychophysics. Hoboken: Wiley. Barter, J. W., Li, S., Sukharnikova, T., Rossi, M. A., Bartholomew, R. A., & Yin, H. H. (2015). Basal ganglia outputs map instantaneous position coordinates during behavior. The Journal of Neuroscience, 35(6), 2703–2716. https://doi.org/ 10.1523/JNEUROSCI.3245-14.2015. Bell, H. C. (2014). Behavioral variability in the service of constancy. International Journal of Comparative Psychology, 27, 338–360. Bell, H. C., & Pellis, S. M. (2011). A cybernetic perspective on food protection in rats: Simple rules can generate complex and adaptable behavior. Animal Behaviour, 82, 659–666. Bennett, S. (1993). A history of control engineering, 1930–1955. London: Peter Peregrinus, Ltd. Bourbon, T. (1990). Invitation to the dance: Explaining the variance when control systems interact. American Behavioral Scientist, 34, 95–105. Carey, T. A. (2006). The method of levels: How to do psychotherapy without getting in the way. Hayward: Living Control Systems. Carey, T. A. (2008). Hold that thought! St. Louis: Newview. Carver, C. S., & Scheier, M. F. (1981). Attention and self-regulation: A control-theory approach to human behavior. New York: Springer. Chapman, S. (1968). Catching a baseball. American Journal of Physics, 36, 868–870. Cohen, B. (1946). Authenticity of scientific anecdotes. Nature, 157, 196–197. Craik, K. J. W. (1947). Theory of the human operator in control systems. British Journal of Psychology, 38, 56–61. Craik, K. J. W. (1948). Theory of the human operator in control systems. British Journal of Psychology, 39, 142–148. Dawkins, R. (1989). The selfish gene (2nd ed.). New York: Oxford University Press. Donders, F. C. (1969). On the speed of mental processes. Acta Psychologica, 30, 412–431. (Attention and performance II (W. G. Koster, ed.). Originally published in 1868).

131

132

Bibliography

Dylan, B. (1963a). Talkin’ WWIII blues. Special Rider Music. Dylan, B. (1963b). The times they are a’changin’. Special Rider Music. Ferster, C. B., & Skinner, B. F. (1957). Schedules of reinforcement. East Norwalk: Appleton-Century-Crofts. Feynman, R. (1985). Surely you’re joking, Mr. Feynman! New York: W. W. Norton. Gourevitch, V. (2018). Of the social contract. In V. Gourevitch (Ed.), The social contract and other later political writings. Cambridge texts in the history of political thought (pp. 39–155, 2nd ed.). Cambridge: Cambridge University Press. Griffin, D. R. (1974). Listening in the dark: The acoustic orientation of bats and men. New York: Dover Publications. Hughes, R. N., Watson, G. D. R., Petter, G. A., Kim, N., Bakhurin, K. I., & Yin, H. H. (2019). Precise coordination of three-dimensional rotational kinematics by ventral tegmental area GABAergic neurons. Current Biology, 29, 1–12. Ioannidis, J. P. (2012). Why science is not necessarily self-correcting. Perspectives on Psychological Science, 7, 645–654. Jagacinski, R., & Flach, J. (2002). Control theory for humans: Quantitative approaches to modeling performance. Hillsdale: Erlbaum. James, W. (1890). The principles of psychology. New York: Henry Holt and Company. (Republished in 1950 by Dover Publications, Inc., Mineola, NY.) Kaufman, L., & Rousseeuw, P. J. (2005). Finding groups in data: An introduction to cluster analysis. Hoboken: Wiley. Keysers, C. (2010). Mirror neurons. Current Biology, 19, 971–973. Kim, N., Barter, J. W., Sukharnikova, T., & Yin, H. H. (2014). Striatal firing rate reflects head movement velocity. The European Journal of Neuroscience, 40(10), 3481–3490. https://doi.org/10.1111/ejn.12722. Kim, N., Li, H. E., Hughes, R. N., Watson, G. D. R., Gallegos, D., West, A. E., Kim, I. H., & Yin, H. H. (2019). A striatal interneuron circuit for continuous target pursuit. Nature Communications, 10, Article number: 2715. Koestler, A. (1990a). The sleepwalkers: A history of man’s changing vision of the universe. London: Penguin Group. Koestler, A. (1990b). The ghost in the machine. London: Penguin Books. Koshland, D. (1980). Behavioral chemotaxis as a model behavioral system. New York: Raven. Labov, W. (1963). The social motivation of a sound change. WORD, 19, 273–309. Liberman, R. P. (1971). Behavior modification with chronic mental patients. Journal of Chronic Diseases, 23, 803–806. Lorenz, K. (1970). Studies in animal and human behaviour (Vol. 1, R. Martin, Trans.). London: Methuen. Lovaas, O. I. (1987). Behavioral treatment and normal educational and intellectual functioning in young autistic children. Journal of Consulting and Clinical Psychology, 55, 3–9. Marken, R. S. (1980). The cause of control movements in a tracking task. Perceptual and Motor Skills, 51, 755–758. Marken, R. S. (1985). Selection of consequences: Adaptive behavior from random reinforcement. Psychological Reports, 56, 379–383.

Bibliography

133

Marken, R. S. (1986). Perceptual organization of behavior: A hierarchical control model of coordinated action. Journal of Experimental Psychology: Human Perception & Performance, 12, 67–76. Marken, R. S. (1989). Behavior in the first degree. In W. Hershberger (Ed.), Volitional action: Conation and control (pp. 299–314). North-Holland: Elsevier. Marken, R. S. (1990). Spreadsheet analysis of a hierarchical control system model of behavior. Behavior Research Methods, Instruments, & Computers, 22, 349–359. Marken, R. S. (1991). Degrees of freedom in behavior. Psychological Science, 2, 92–100. Marken, R. S. (1999). PERCOLATe: Perceptual Control Analysis of Tasks. International Journal of Human-Computer Studies, 50, 481–487. Marken, R. S. (2001). Controlled variables: Psychology as the center fielder views it. American Journal of Psychology, 114, 259–281. Marken, R. S. (2002a). Looking at behavior through control theory glasses. Review of General Psychology, 6, 260–270. Marken, R. S. (2002b). Hierarchical behavior of perception. In R. Marken (Ed.). More mind readings: Methods and models in the study of purpose (pp. 85–112). Chapel Hill: Newview. Marken, R. S. (2005a). Optical trajectories and the informational basis of fly ball catching. Journal of Experimental Psychology: Human Perception & Performance, 31(3), 630–634. Marken, R. S. (2005b). A model-based approach to prioritizing medical safety practice. In K. Henriksen, J. Battles, E. Marks, & D. Lewin (Eds.), Advances in patient safety: From research to implementation. Vol. 2, Concepts and methodology. AHRQ Publication No. 05-0-21-2 (pp. 409–424). Rockville: Agency for Healthcare Research and Quality. Marken, R. S. (2013). Testing for controlled variables: A model-based approach to determining the perceptual basis of behavior. Attention, Perception and Psychophysics, 76, 255–263. Marken, R. S. (2014). Doing research on purpose: A control theory approach to experimental psychology. St. Louis: Newview. Marken, R. S. (2020). Understanding purposeful systems: A comparison of the engineering and psychological approaches to control. In W. Mansell (Ed.), The interdisciplinary handbook of perceptual control theory (pp 9–23). Holland: Elsevier. Marken, R. S., & Carey, T. A. (2014). Understanding the change process involved in solving psychological problems: A model-based approach to understanding how psychotherapy works. Clinical Psychology & Psychotherapy, 22, 580–590. https://doi.org/10.1002/cpp.1919. Marken, R. S., & Carey, T. A. (2015). Controlling people: The paradoxical nature of being human. Samford Valley: Australian Academic Press. Marken, R. S., & Horth, B. (2011). When causality does not imply correlation: More spadework at the foundations of scientific psychology. Psychological Reports, 108, 1–12. Marken, R. S., Mansell, W., & Khatib, A. (2013). Motor control as the control of perception. Perceptual and Motor Skills, 117, 236–247.

134

Bibliography

Marken, R. S., & Powers, W. T. (1989a). Levels of intention in behavior. In W. Hershberger (Ed.), Volitional action: Conation and control (pp. 299–314). NorthHolland: Elsevier. Marken, R. S., & Powers, W. T. (1989b). Random-walk chemotaxis: Trial-anderror as a control process. Behavioral Neuroscience, 103, 1348–1355. McBeath, M. K., Shaffer, D. M., & Kaiser, M. K. (1995). How baseball outfielders determine where to run to catch fly balls. Science, 268, 569–573. McClelland, K. (2004). The collective control of perceptions: Constructing order from conflict. International Journal of Human-Computer Studies, 60, 65–99. McPhail, C., Powers, W. T., & Tucker, C. W. (1992). Simulating individual and collective action in temporary gatherings. Social Science Computer Review, 10, 1–28. McSweeney, F. K., & Murphy, E. S. (2014). The Wiley Blackwell handbook of operant and classical conditioning. New York: Wiley. Mechsner, F., Kerzel, K., Knoblich, G., & Prinz, W. (2001). Perceptual basis of bimanual coordination. Nature, 414, 69–73. Miller, G. A., & Heise, G. A. (1950). The trill threshold. Journal of the Acoustical Society of America, 22, 637–638. Neisser, U. (1967). Cognitive psychology. Englewood Cliffs: Prentice-Hall. Pavlov, I. P. (1928). Lectures on conditioned reflexes (W. H. Gantt, Trans.). London: Allen and Unwin. Petrie, H. G. (1981). The dilemma of enquiry and learning. Chicago: University of Chicago Press. Piaget, J., & Cook, M. T. (1952). The origins of intelligence in children. New York: International University Press. Pierce, D. W., & Cheney, C. D. (2013). Behavior analysis and learning. New York: Psychology Press. Plooij, F. X. (1987). Infant-ape behavioral development, the control of perception, types of learning and symbolism. In J. Montenegro, A. Tryphon, & S. Dionnet (Eds.), Symbolism and knowledge (pp. 33–64). Geneva: Archives Jean Piaget Foundation. Plooij, F. X. (2013). The trilogy of mind. In M. Heimann (Ed.), Regression periods in human infancy (pp. 185–205). Mahwah: Erlbaum. Plooij, F. X., & van de Rijt-Plooij, H. H. C. (1990). Developmental transitions as successive reorganizations of a control hierarchy. American Behavioral Scientist, 34, 67–80. Powers, W. T. (1971). A feedback model for behavior: Analysis of a rat experiment. Behavioral Science, 16, 558–563. Powers, W. T. (1973a). Feedback: Beyond behaviorism. Science, 179, 351–356. Powers, W. T. (1973b). Behavior: The control of perception. Chicago: Aldine De Gruyter. Powers, W. T. (1978). Quantitative analysis of purposive systems: Some spadework at the foundations of experimental psychology. Psychological Review, 85, 417–435. Powers, W. T. (1979a). The nature of robots: Part 3: A closer look at human behavior. BYTE, August, p. 109. Powers, W. T. (1979b). The nature of robots: Part 4: Testing for controlled variables. BYTE, September.

Bibliography

135

Powers, W. T. (1979c). A cybernetic model for research in human development. In M. Ozer (Ed.), A cybernetic approach to the assessment of children: Toward a more humane use of human beings (pp. 11–66). New York: Routledge. Powers, W. T. (1980). A systems approach to consciousness. In R. J. Davidson & J.  M. Davidson (Eds.), The psychobiology of consciousness (pp. 217–242). New York: Plenum Press. Powers, W. T. (1985). Comment on the “phantom plateau”. Perceptual and Motor Skills, 61, 329–330. Powers, W. T. (1989) Quantitative measurement of volition. In W. Hershberger (Ed.). Volitional Action: Conation and control (pp. 315–332). North-Holland: Elsevier Science Publishers. Powers, W. T. (1990). Control theory and statistical generalizations. American Behavioral Scientist, 34, 24–31. Powers, W. T. (1992). Post to Control Systems Group Network (CSGnet). CSG-l@ UIUCVMD.BITNET, June 25, 1992. Powers, W. T. (2005a). Living control systems III: The fact of control. Bloomfield, NJ: Benchmark Publications. Powers, W. T. (2005b). Behavior: The control of perception (2nd ed.). Bloomfield: Benchmark Publications. Powers, W. T., Clark, R. K., & McFarland, R. L. (1960a). A general feedback theory of behavior: Part I. Perceptual and Motor Skills, 11, 71–88. Powers, W. T., Clark, R. K., & McFarland, R. L. (1960b). A general feedback theory of behavior: Part II. Perceptual and Motor Skills, 11, 309–323. Premack, D., & Premack, A. (1984). The mind of an ape. New York: W. W. Norton. Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioral model. ACM SIGGRAPH Computer Graphics, 21, 25–34. Richard, P. B. (1983). Mechanisms and adaptation in the constructive behaviour of the beaver (C. fiber L.). Acta Zoologica Fennica, 174, 105–108. Rijt-Plooij, H. H. C., & van de and Plooij, F. X. (1992). Infantile regressions: Disorganization and the onset of transition periods. Journal of Reproductive and Infant Psychology, 10(3), 129–149. Robertson, R., & Glines, L. A. (1985). The phantom plateau returns. Perceptual and Motor Skills, 61, 55–64. Rosenbaum, D. A. (1987). Hierarchical organization of motor programs. In S. P. Wise (Ed.), Higher brain functions: Recent explorations of the brain’s emergent properties (pp. 45–66). New York: Wiley. Rosenblueth, A., Wiener, N., & Bigelow, J. (1943). Behavior, purpose and teleology. Philosophy of Science, 10, 18–24. Runkel, P. J. (1990). Casting nets and testing specimens. New York: Praeger. Runkel, P. J. (2003). People as living things: The psychology of perceptual control. Hayward: Living Control Systems Publishing. Scholte, T. (2019). Modeling workplace conflict with systems theatre. Journal on Systemics, Cybernetics and Informatics, 17, 31–38. Shaffer, D. M., Krauchunas, S. M., Eddy, M., & McBeath, M. K. (2004). How dogs navigate to catch frisbees. Psychological Science, 15, 437–441.

136

Bibliography

Shaffer, D. M., Marken, R. S., Dolgov, I., & Maynor, A. B. (2013). Chasin’ choppers: Using unpredictable trajectories to test theories of object interception. Attention, Perception and Psychophysics, 75, 1496–1506. Shaffer, D. M., Marken, R. S., Dolgov, I., & Maynor, A. B. (2015). Catching objects thrown to oneself: Testing the generality of a control strategy for object interception. Perception, 44, 400–409. Shepard, R. N., & Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171, 701–703. Sheridan, T. B., & Ferrell, W. R. (1974). Man-machine systems: Information, control, and decision models of human performance. Cambridge: MIT Press. Skinner, B. F. (1971). Beyond freedom and dignity. Indianapolis: Hackett Publishing. Sternberg, S. (1969). The discovery of processing stages: Extensions of Donders’ method. Acta Psychologica, 30, 276–315. Tiwari, M. (2011). Science behind human saliva. Journal of Natural Science, Biology, and Medicine, 2, 53–58. https://doi.org/10.4103/0976-9668.82322. Tolman, E. C. (1922). A new formula for behaviorism. Psychological Review, 29, 44–53. Turing, A. (1948). Intelligent machinery. Reprinted in Turing, A. M. (1996). “Intelligent machinery, a heretical theory”. Philosophia Mathematica, 4(3), 256–260. https://doi.org/10.1093/philmat/4.3.256. Turvey, M. T., Shaw, R. E., & Mace, N. (1978). Issues in the theory of action: Degrees of freedom, coordinative structures and coalitions. In J. Requin (Ed.), Attention and performance VII (pp. 557–595). Hillsdale: Erlbaum. Vanderijt, H., & Plooij, F. (2003). The wonder weeks. Emmaus, Pennsylvania: Rodale. Verhave, T. (1959). Technique for differential reinforcement of rate of avoidance responding. Science, 129, 959–960. Warren, R. M., Obusek, C. J., Farmer, R. M., & Warren, R. P. (1969). Auditory sequence: Confusion of patterns other than speech or music. Science, 164, 586–587. Weimerskirch, H., Martin, J., Clerquin, Y., Alexandre, P., & Jiraskova, S. (2001). Energy saving in flight formation. Nature, 413, 697–698. Wickens, C. D., Hollands, J. G., Parasuraman, R., & Banbury, S. (2012). Engineering psychology and human performance (4th ed.). Upper Saddle River: Pearson. Willett, A. B. S., Marken, R. S., Parker, M. G., & Mansell, W. (2017). Control blindness: Why people can make incorrect inferences about the intentions of others. Attention, Perception & Performance, 79, 841–849. https://doi.org/10.3758/ s13414-016-1268-3. Yin, H. (2013). Restoring purpose to behavior. In G. Baldassarre & M. Mirolli (Eds.), Computational and robotic models of the hierarchical organization of behavior (pp. 319–347). Berlin: Springer-Verlag. Yin, H. H. (2016). The basal ganglia and hierarchical control in voluntary behavior. In J.-J. Soghomonian (Ed.), The basal ganglia (pp. 513–566). New York: Springer. Yin, H. H. (2017). The basal ganglia in action. Neuroscientist, 23, 299–313.

Index

agreement, to cooperative control, 111 Asch conformity study, 38 Asch, Solomon, 38 awareness, 99

contingent relationship, 72 control objectivized side effect of, 9 of perception, 6 reorganization and parameters of, 101 control blindness, 11 control loop variables, 48 control of perception, control theory, 3, 37 controlled conditions, 28 controlled perception, 1, 39 Test for Controlled Variable and, 23 controlled variables, x, 4, 9 behavioral illusion, 14 estimation of, 23 hierarchical control, 76 hierarchical organization and, 123 hypotheses of, 41 testing for, 53–55 two-dimensional control, 55 virtual, 114 controllers, in manipulative control, 120 cooperative control, 104–113, 110 correlation limitations of, 60 0.997, 45–46 CROWD program, 106 culture, cooperation and, 108

behavior carrot stick approach, 118 control of perception and, 123 hierarchical control and, 82 identification of, 36 manipulative control of, 117 purposes of, behavioral illusion, 13 behavioral results, 8 Bourbon, Tom, 105 Brahe, Tycho, 16 carrot stick approach, 118 casting nets data collection, 42–46 cause–effect model, 50 centralization index (CI), 108 Chapman, S., 63 classical conditioning, 37 closed-loop relationship, 4 cognition, 79 Coin Game, 21–22 compensatory tracking, 48 locus of control and, 50 operant analog, 57 reorganization research and, 101 two-dimensional control, 55 completely randomized design, 43 compliance behavior, 38 computer-simulated control systems, 43 configuration perceptions, 77 conflictive control characteristics of, 113 strength systems of, 115 confounding variables, 28 confusions, S-R error and, 11 consciousness, reorganization and, 99, 102

data collection and analysis, 42–46, 122 classification and, 126 dead zone, in conflictive control, 115 dependent variable independent variable and, 39 manipulation of, 28 disturbance effect, 49 disturbance-resistance approach, 6, 11, 117 E. coli reorganization, 91 egg-rolling behavior study, 41 electromagnetic induction, 33

137

138

Index

emotion, PCT and research on, 129 endurance, in control systems, 116 environment equation, 5, 58 error signal, 9 ethology, reorganization research and, 97 event perceptions, 78 evolution, cooperative control and, 111 face value, limits of, 16 Faraday, Michael, 33 feedback, 5 Feynman, Richard, 23 fixed action pattern, 42 fixed reference, 24 Glines, L. A., 92, 94, 95, 98, 99, 100, 102 goal-oriented behavior, 2 Goodall, Jane, 97 goodness of fit, 60 group cooperation, 106 hierarchical control behavioral evidence for, 82 laboratory tests of, 73 levels of, 81 of perception, 68, 123 perceptual types and, 86 portable demonstrator test, 71, 127 variables in, 76 hypothesis of controlled perception, 19 of controlled variable, 41 imitation, 108 independent variable dependent variable and, 39 manipulation of, 28 intensity perceptions, 76 interpersonal conflict, 113, 116 intrapersonal conflict, 113 intrinsic state, 90 intrinsic variables, 90 James, William, 2 Kepler, Johannes, 16 laboratory testing, hierarchical control, 73 Labov, W., 108 learning tasks, reorganization research and, 92 living control system future research on, 128 locus of control, 50 loop gain, 7

manipulative control, 117 carrot stick approach, 118 controllers and, 120 man–machine blunder, 9 Manual Control Theory (MCT), 5–7 meme, 108 mind modeling, 29 Mind Reading (computer demo), 24 mind reading, TCV as, 23–27 mirror neuron, 111 motor outputs, 80 naturalistic observation control research and, 97 negative feedback, 5 nested systems hierarchical control, 71 reorganization theory and, 95 neural control systems, 81 Newton, Isaac, 16 Ohm's law, 46 one-dimensional tracking, 56 operant conditioning, 58, 61 operational definition, 41 optical acceleration (OAC) hypothesis, 63 participants in research, 42–46 perception cognition and, 79 control of, 11, 68 hypothesis of control of, 19 of perceiving, 99 perceptual control equations, 6 Perceptual Control Theory (PCT), x, 5–7 applications for, 32–34 Coin Game and, 21 conflictive control and, 116 data gathering for, 122 emotion and, 129 future research issues, 122 hypothetical levels of, 76 locus of control and, 50 manipulative control, 117 operant conditioning and, 58 overt behavior and, 16 real world cases, 63–66 reorganization of control systems, 89 perceptual results, 8 perceptual types, 86 perceptual variables, 9 behavioral evidence for, 82 controlled perception and, 39 hierarchical control, 70 hierarchy of, 123

Index of purpose, 18 reorganization theory of learning and, 92 types of, 40 Piaget, Jean, 126 Plooij, Frans, 97 points of view, 100 portable demonstrator test, 127 Powers, William T., 9, 10, 43, 48 living control system research and, 128 locus of control and, 50 on manipulative control, 118 reorganization research and, 101 principle perceptions, 79 program perceptions, 78 Psychological Methods, ix psychological research, control theory and, 122 psychophysical method of adjustment, 83 psychotherapy, reorganization in, 96 purpose basic research on, 48 future research on, 129 perceptual variables of, 18 purposeful behavior, pursuit tracking, 30 random ratio, 61 random-walk process, 91 ratio schedule, 61 reference signal, 8, 52 centralization index and, 109 reference state, 4, 36 perceptual variables and, 39 virtual state, 114 regression periods control theory and, 97 reinforcement, behavior and, 61 relationship perceptions, 78 reorganization compensatory tracking and, 101 consciousness and, 99 control systems and, 89 ethological approach to, 97 evidence of, 91 in psychotherapy, 96 repeated measures design, 43 rewards, control systems and, 44

139

Reynolds, Craig, 106 Rijt-Plooij, Hetty van de, 97 robbing and dodging behavior, 115 Robertson, R., 92, 94, 95, 98, 99, 100, 102 root mean square (rms) deviation, 60 Rousseau, Jean-Jacques, 112 rubber band demo, 12, 117 runaway condition, 73 Runkel, P. J., 28 salivation conditioning studies, 37 schedules of reinforcement, 61 scientifically significant results, 45–46 sensation perceptions, 77 sensory and motor nuclei, 81 sequence perceptions, 78 shock avoidance research, 60 social contract, 112 social control, 104 speed of response, 7 spinal reflex, 81 S-R error, 11–13 stability factor, 54 statistically significant results, 45–46 stimulus–response behavioral theory, 8, 37 striatal interneurons, 82 surveys, controlled variables in, 39 system equation, 5, 58 system perception, 79 test for the controlled variable (TCV), 19–23 applications for, 31–34 mind reading and, 23–27 procedures and steps in, 23–32 reorganization research and, 97 transfer function, 8 transition perceptions, 78 Turing Machine, 80 two-dimensional control, 55 two-person cooperation, 105 Verhave, T., 58 vertical optical velocity, 66 volition, 99 Willett, A. B. S., 11