Anatomy of the Mind Exploring Psychological Mechanisms and Processes

Views:
 
Category: Education
     
 

Presentation Description

This book aims to understand human cognition and psychology through a comprehensive computational theory of the human mind, namely, a computational "cognitive architecture" (or more specifically, the Clarion cognitive architecture). The goal of this work is to develop a unified framework for understanding the human mind, and within the unified framework, to develop process-based, mechanistic explanations of a large variety of psychological phenomena. Specifically, the book first describes the essential Clarion framework and its cognitive-psychological justifications, then its computational instantiations, and finally its applications to capturing, simulating, and explaining various psychological phenomena and empirical data. The book shows how the models and simulations shed light on psychological mechanisms and processes through the lens of a unified framework. In fields ranging from cognitive science, to psychology, to artificial intelligence, and even to philosophy, researchers, graduate and undergraduate students, and practitioners of various kinds may have interest in topics covered by this book. The book may also be suitable for seminars or courses, at graduate or undergraduate levels, on cognitive architectures or cognitive modeling (i.e. computational psychology).

Comments

Presentation Transcript

slide 2:

Anatomy of the Mind

slide 3:

OXFORD SERIES ON COGNITIVE MODELS AND ARCHITECTURES Series Editor Frank E. Ritter Series Board Rich Carlson Gary Cottrell Robert L. Goldstone Eva Hudlicka Pat Langley Robert St. Amant Richard M. Y oung Integrated Models of Cognitive Systems Edited by Wayne D. Gray In Order to Learn: How the Sequence of Topics Infuences Learning Edited by Frank E. Ritter Joseph Nerb Erno Lehtinen and Timothy O’Shea How Can the Human Mind Occur in the Physical Universe By John R. Anderson Principles of Synthetic Intelligence PSI: An Architecture of Motivated Cognition By Joscha Bach The Multitasking Mind By David D. Salvucci and Niels A. Taatgen How to Build a Brain: A Neural Architecture for Biological Cognition By Chris Eliasmith Minding Norms: Mechanisms and Dynamics of Social Order in Agent Societies Edited by Rosaria Conte Giulia Andrighetto and Marco Campennì Social Emotions in Nature and Artifact Edited by Jonathan Gratch and Stacy Marsella Anatomy of the Mind: Exploring Psychological Mechanisms and Processes with the Clarion Cognitive Architecture By Ron Sun

slide 4:

1 Anatomy of the Mind Exploring Psychological Mechanisms and Processes with the Clarion Cognitive Architecture Ron Sun

slide 5:

1 Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research scholarship and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and certain other countries. Published in the United States of America by Oxford University Press 198 Madison Avenue New York NY 10016 United States of America. © Oxford University Press 2016 First Edition published in 2016 All rights reserved. No part of this publication may be reproduced stored in a retrieval system or transmitted in any form or by any means without the prior permission in writing of Oxford University Press or as expressly permitted by law by license or under terms agreed with the appropriate reproduction rights organization. Inquiries concerning reproduction outside the scope of the above should be sent to the Rights Department Oxford University Press at the address above. You must not circulate this work in any other form and you must impose this same condition on any acquirer. Library of Congress Cataloging-in-Publication Data Sun Ron 1960– Anatomy of the mind : exploring psychological mechanisms and processes with the Clarion cognitive architecture / Ron Sun. pages cm. — Oxford series on cognitive models and architectures Includes bibliographical references and index. ISBN 978–0–19–979455–3 1. Cognitive science. 2. Cognitive neuroscience. 3. Computer architecture. 4. Cognition—Computer simulation. I. Title. BF311.S8148 2016 153—dc23 2015018557 9 8 7 6 5 4 3 2 1 Printed by Sheridan USA

slide 6:

v Contents Preface  xiii 1. What is A Cognitive Architecture  1 1.1. A Theory of the Mind and Beyond  1 1.2. Why Computational Models/Theories  3 1.3. Questions about Computational Models/Theories  7 1.4. Why a Computational Cognitive Architecture  9 1.5. Why Clarion  13 1.6. Why This Book  15 1.7. A Few Fundamental Issues  16 1.7.1. Ecological-Functional Perspective  16 1.7.2 Modularity  17 1.7.3. Multiplicity of Representation  18 1.7.4. Dynamic Interaction  19 1.8. Concluding Remarks  20 2. Essential Structures of the Mind  21 2.1. Essential Desiderata  21 2.2. An Illustration of the Desiderata  24 2.3. Justifying the Desiderata  26 2.3.1. Implicit-Explicit Distinction and Synergistic Interaction  27 2.3.2. Separation of the Implicit-Explicit and the Procedural-Declarative Distinction  30

slide 7:

vi Contents 2.3.3. Bottom-Up and Top-Down Learning  34 2.3.4. Motivational and Metacognitive Control  36 2.4. Four Subsystems of Clarion  37 2.4.1. Overview of the Subsystems  37 2.4.2. The Action-Centered Subsystem  40 2.4.3. The Non-Action-Centered Subsystem  42 2.4.4. The Motivational Subsystem  43 2.4.5. The Metacognitive Subsystem  44 2.4.6. Parameters of the Subsystems  45 2.5. Accounting for Synergy within the Subsystems of Clarion 45 2.5.1. Accounting for Synergy within the ACS  46 2.5.2. Accounting for Synergy within the NACS  48 2.6. Concluding Remarks  50 3. The Action-Centered and Non-Action-Centered Subsystems  51 3.1. The Action-Centered Subsystem  52 3.1.1. Background  52 3.1.2. Representation  54 3.1.2.1. Representation in the Top Level  54 3.1.2.2. Representation in the Bottom Level  57 3.1.2.3. Action Decision Making  57 3.1.3. Learning  63 3.1.3.1. Learning in the Bottom Level  63 3.1.3.2. Learning in the Top Level  65 3.1.4. Level Integration  67 3.1.5. An Example  68 3.2. The Non-Action-Centered Subsystem  69 3.2.1. Background  69 3.2.2. Representation  72 3.2.2.1. Overall Algorithm  72 3.2.2.2. Representation in the Top Level  73 3.2.2.3. Representation in the Bottom Level  77 3.2.2.4. Representation of Conceptual Hierarchies  81 3.2.3. Learning  81 3.2.3.1. Learning in the Bottom Level  81 3.2.3.2. Learning in the Top Level  82 3.2.4. Memory retrieval  83 3.2.5. An Example  85

slide 8:

Contents vii 3.3. Knowledge Extraction Assimilation and T ransfer  87 3.3.1. Background  87 3.3.2. Bottom-Up Learning in the ACS  88 3.3.2.1. Rule Extraction and Refnement  88 3.3.2.2. Independent Rule Learning  93 3.3.2.3. Implications of Bottom-Up Learning  94 3.3.3. Top-Down Learning in the ACS  96 3.3.4. Transfer of Knowledge from the ACS to the NACS  97 3.3.5. Bottom-Up and Top-Down Learning in the NACS  100 3.3.6. Transfer of Knowledge from the NACS to the ACS  101 3.3.7. An Example  101 3.3.7.1. Learning about “Knife”  102 3.3.7.2. Learning about “Knife” within Clarion  103 3.3.7.3. Learning More Complex Concepts within Clarion  106 3.4. General Discussion  108 3.4.1. More on the Two Levels  108 3.4.2. More on the Two Learning Directions  110 3.4.3. Controversies  112 3.4.4. Summary  113 Appendix: Additional Details of the ACS and the NACS  113 A.1. Response Time  113 A.1.1. Response Time of the ACS  113 A.1.2. Response Time of the NACS  115 A.2. Learning in MLP Backpropagation Networks  116 A.3. Learning in Auto-Associative Networks  117 A.4. Representation of Conceptual Hierarchies  118 4. The Motivational and Metacognitive Subsystems  121 4.1. Introduction  121 4.2. The Motivational Subsystem  123 4.2.1. Essential Considerations  123 4.2.2. Drives  126 4.2.2.1. Primary Drives  126 4.2.2.2. Secondary Drives  129 4.2.2.3. Approach versus Avoidance Drives  130 4.2.2.4. Drive Strengths  131

slide 9:

viii Contents 4.2.3. Goals  132 4.2.4. Modules and Their Functions  133 4.2.4.1. Initialization Module  133 4.2.4.2. Preprocessing Module  134 4.2.4.3. Drive Core Module  134 4.2.4.4. Defcit Change Module  135 4.3. The Metacognitive Subsystem  135 4.3.1. Essential Considerations  136 4.3.2. Modules and Their Functions  137 4.3.2.1. Goal Module  137 4.3.2.2. Reinforcement Module  140 4.3.2.3. Processing Mode Module  141 4.3.2.4. Input/Output Filtering Modules  143 4.3.2.5. Reasoning/Learning Selection Modules  144 4.3.2.6. Monitoring Buffer  145 4.3.2.7. Other MCS Modules  145 4.4. General Discussion  146 4.4.1. Reactivity versus Motivational Control  146 4.4.2. Scope of the MCS  146 4.4.3. Need for the MCS  148 4.4.4. Information Flows Involving the MS and the MCS  148 4.4.5. Concluding Remarks  149 Appendix: Additional Details of the MS and the MCS  149 A.1. Change of Drive Defcits  149 A.2. Determining Avoidance versus Approach Drives Goals and Behaviors  150 A.3. Learning in the MS  151 A.4. Learning in the MCS  153 A.4.1. Learning Drive-Goal Connections  153 A.4.2. Learning New Goals  154 5. Simulating Procedural and Declarative Processes  155 5.1. Modeling the Dynamic Process Control Task  157 5.1.1. Background  157 5.1.2. Task and Data  158 5.1.3. Simulation Setup  160

slide 10:

Contents ix 5.1.4. Simulation Results  162 5.1.5. Discussion  166 5.2. Modeling the Alphabetic Arithmetic Task  168 5.2.1. Background  168 5.2.2. Task and Data  169 5.2.3. Top-Down Simulation  171 5.2.3.1. Simulation Setup  171 5.2.3.2. Simulation Results  174 5.2.4. Alternative Simulations  178 5.2.5. Discussion  181 5.3. Modeling the Categorical Inference Task  183 5.3.1. Background  183 5.3.2. Task and Data  185 5.3.3. Simulation Setup  187 5.3.4. Simulation Results  190 5.3.5. Discussion  192 5.4. Modeling Intuition in the Discovery Task  194 5.4.1. Background  194 5.4.2. Task and Data  195 5.4.3. Simulation Setup  198 5.4.4. Simulation Results  200 5.4.5. Discussion  203 5.5. Capturing Psychological “Laws”  205 5.5.1. Uncertain Deductive Reasoning  205 5.5.1.1. Uncertain Information  206 5.5.1.2. Incomplete Information  206 5.5.1.3. Similarity  207 5.5.1.4. Inheritance  207 5.5.1.5. Cancellation of Inheritance  208 5.5.1.6. Mixed Rules and Similarities  208 5.5.2. Reasoning with Heuristics  209 5.5.2.1. Representativeness Heuristic  209 5.5.2.2. Availability Heuristic  212 5.5.2.3. Probability Matching  214 5.5.3. Inductive Reasoning  215 5.5.3.1. Similarity between the Premise and the Conclusion  215 5.5.3.2. Multiple Premises  216 5.5.3.3. Functional Attributes  217

slide 11:

x Contents 5.5.4. Other Psychological “Laws”  218 5.5.5. Discussion of Psychological “Laws”  220 5.6. General Discussion  221 6. Simulating Motivational and Metacognitive Processes  225 6.1. Modeling Metacognitive Judgment  225 6.1.1. Background  225 6.1.2. Task and Data  226 6.1.3. Simulation Setup  227 6.1.4. Simulation Results  228 6.1.5. Discussion  229 6.2. Modeling Metacognitive Inference  229 6.2.1. Task and Data  229 6.2.2. Simulation Setup  230 6.2.3. Simulation Results  232 6.2.4. Discussion  232 6.3. Modeling Motivation-Cognition Interaction  234 6.3.1. Background  234 6.3.2. Task and Data  238 6.3.3. Simulation Setup  241 6.3.4. Simulation Results  244 6.3.5. Discussion  246 6.4. Modeling Human Personality  247 6.4.1. Background  247 6.4.2. Principles of Personality Within Clarion  250 6.4.2.1. Principles and Justifcations  250 6.4.2.2. Explaining Personality  254 6.4.3. Simulations of Personality  258 6.4.3.1. Simulation 1  258 6.4.3.2. Simulation 2  263 6.4.3.3. Simulation 3  267 6.4.4. Discussion  271 6.5. Accounting for Human Moral Judgment  272 6.5.1. Background  272 6.5.2. Human Data  275 6.5.2.1. Effects of Personal Physical Force  275 6.5.2.2. Effects of Intention  276 6.5.2.3. Effects of Cognitive Load  276

slide 12:

Contents xi 6.5.3. Two Contrasting Views  277 6.5.3.1. Details of Model 1  278 6.5.3.2. Details of Model 2  279 6.5.4. Discussion  281 6.6. Accounting for Human Emotion  283 6.6.1. Issues of Emotion  283 6.6.2. Emotion and Motivation  284 6.6.3. Emotion and the Implicit-Explicit Distinction  285 6.6.4. Effects of Emotion  286 6.6.5. Emotion Generation and Regulation  287 6.6.6. Discussion  289 6.7. General Discussion  289 7. Cognitive Social Simulation  293 7.1. Introduction and Background  293 7.2. Cognition and Survival  295 7.2.1. Tribal Society Survival Task  295 7.2.2. Simulation Setup  297 7.2.3. Simulation Results  300 7.2.3.1. Effects of Social and Environmental Factors  300 7.2.3.2. Effects of Cognitive Factors  302 7.2.4. Discussion  307 7.3. Motivation and Survival  309 7.3.1. Simulation Setup  309 7.3.2. Simulation Results  314 7.3.2.1. Effects of Social and Environmental Factors  314 7.3.2.2. Effects of Cognitive Factors  315 7.3.2.3. Effects of Motivational Factors  318 7.3.3. Discussion  320 7.4. Organizational Decision Making  322 7.4.1. Organizational Decision Task  322 7.4.2. Simulations and Results  325 7.4.2.1. Simulation I: Matching Human Data  325 7.4.2.2. Simulation II: Extending Simulation Temporally  326 7.4.2.3. Simulation III: Varying Cognitive Parameters  329

slide 13:

xii Contents 7.4.2.4. Simulation IV: Introducing Individual Differences  332 7.4.3. Discussion  333 7.5. Academic Publishing  334 7.5.1. Academic Science  334 7.5.2. Simulation Setup  336 7.5.3. Simulation Results  338 7.5.4. Discussion  342 7.6. General Discussion  343 7.6.1. Theoretical Issues in Cognitive Social Simulation  343 7.6.2. Challenges  346 7.6.3. Concluding Remarks  347 8. Some Important Questions and Their Short Answers  349 8.1. Theoretical Questions  349 8.2. Computational Questions  367 8.3. Biological Connections  378 9. General Discussions and Conclusions  381 9.1. A Summary of the Cognitive Architecture  381 9.2. A Discussion of the Methodologies  383 9.3. Relations to Some Important Notions  385 9.4. Relations to Some Existing Approaches  390 9.5. Comparisons with Other Cognitive Architectures  393 9.6. Future Directions  399 9.6.1. Directions for Cognitive Social Simulation  399 9.6.2. Other Directions for Cognitive Architectures  401 9.6.3. Final Words on Future Directions  403 References  405 Index  429

slide 14:

xiii Preface This book aims to understand psychological cognitive mechanisms processes and functionalities through a comprehensive computational theory of the human mind namely a computational “cognitive architec- ture” or more specifcally the Clarion cognitive architecture. The goal of this work is to develop a unifed framework for understanding the human mind and within the unifed framework to develop process-based mech - anistic explanations of a substantial variety of psychological phenomena. The book describes the essential Clarion framework its cognitive- psychological justifcations its computational instantiations and its applications to capturing simulating and explaining various psycholog- ical phenomena and empirical data. The book shows how the models and simulations shed light on psychological mechanisms and processes through the lens of a unifed framework namely Clarion. While a forthcoming companion volume to this book will fully describe the technical details of Clarion along with hands-on examples the present book concentrates more on a conceptual-level exposition and explanation but also describes in a more accessible way essential techni- cal details of Clarion. It covers those technical details that are necessary for explaining the psychological phenomena discussed in this book. The following may be considered the features of the present book: • A scope broader than any other cognitive architecture pointing to new possibilities for developing comprehensive computa- tional cognitive architectures.

slide 15:

xiv Preface • Integration of multiple approaches and perspectives within this broad scope. • Exploration of empirical data and phenomena through com- putational models and simulations examining a variety of data from a variety of empirical felds. • Balance of formal modeling and readability i.e. accessibility to a multidisciplinary readership. These features were designed with potential readers of the book in mind who may include in no particular order: 1 cognitive scien- tists especially cognitive modeling researchers or “computational psy- chologists” as one might call them who might be interested in a new theoretical framework a new generic computational model as well as new interpretations of data through computational modeling 2 exper- imental psychologists who might be interested in new possibilities of interpreting empirical data within a unifed framework new conceptual interpretations or existing interpretations for that matter being sub- stantiated through computational modeling and also new possibilities for further empirical explorations 3 researchers from adjacent felds who might be interested in work on computational psychology cog- nitive modeling and how such research may shed light on the mind 4 interested lay readers who might want to explore computational psychology and its implications for understanding the human mind … and so on. To put it simply this book is for those who are interested in exploring and understanding the human mind through computational models that capture and explain empirical data and phenomena in a unifed framework. In felds ranging from cognitive science especially cognitive mod - eling to psychology to artifcial intelligence and even to philosophy academic researchers graduate and undergraduate students and practi- tioners of various kinds may have interest in topics covered by this book. The book may be suitable for graduate-level seminars or courses on cog- nitive architectures or cognitive modeling but might also be suitable for the advanced undergraduate level. A little history is in order here. The general ideas of a pair of books this one and a companion technical book on Clarion were drawn up in February 2009 after much rumination. I worked more on the ideas for the two books in May of that year. In November between two trips I wrote two book proposals. They were submitted to Oxford University

slide 16:

Preface xv Press in January 2010. After a round of very thorough reviews of the book proposals by the publisher the contracts for the two books were signed in May 2010. The writing of this book was sporadic and largely put off until the summer of 2011. Since that time efforts were made to fnish the book. The manuscript was sent to the publisher at the end of 2013. The history of the Clarion cognitive architecture started of course much earlier than that. Back in the summer of 1994 the ONR cognitive science basic research program issued a call for proposals which prompted me to put together a set of ideas that had been brewing in my head. That was the beginning of Clarion. The grant from the ONR program enabled the development and the validation of the initial version of Clarion. During the 1998–1999 academic year I had my sabbatical leave at the NEC Research Institute. A theoretically oriented book on Clarion took shape during that period which was subsequently published. Starting in 2000 research grants from ARI enabled the further development of a number of subsystems within Clarion. Then from 2008 on new grants from ONR enabled the extension of the work to social simulation and other related topics. I would like to thank Frank Ritter for his solicitation of thorough reviews of the two book proposals and for his suggestions regarding the organizations of the books. Thanks also go to the eight reviewers of the book proposals for their helpful suggestions. Later I received detailed critiques of the entire book manuscript from Frank Ritter and two anonymous reviewers whom I gratefully acknowledge as well. I would also like to acknowledge useful discussions that I have had with many colleagues including Paul Bello Michael Zenzen Larry Reid Jeff White Jun Zhang and Deliang Wang regarding motivation emotion personality ethics learning modeling and so on. I am also indebted to my many collaborators past and present including Sebastien Helie Bob Mathews Sean Lane Selmer Bringsjord Michael Lynch and their students. I also want to acknowledge my past and current graduate students: Jason Xi Zhang Isaac Naveh Nick Wilson Pierson Fleischer and others. Some other students contributed to the work on Clarion as well. The work described in this book is theirs as well as mine. Clarion has been implemented as Java and C libraries available at courtesy of Nick Wilson and Michael Lynch: http://www.clarioncognitivearchitecture.com

slide 17:

xvi Preface Finally the work described here has been fnancially supported in part by ONR grants N00014-95-1-0440 N00014 ‐08 ‐1 ‐0068 and N00014-13-1-0342 as well as ARI grants DASW01-00-K-0012 and W74V8H-05-K-0002. Without these forms of support this work could not have come into being. Ron Sun Troy New York

slide 18:

Anatomy of the Mind

slide 20:

1 1 What Is A Cognitive Architecture In this chapter as an introduction to what is to be detailed in this book I will attempt to justify the endeavor of developing a generic computa- tional model theory of the mind i.e. a computational cognitive archi- tecture through addressing a series of questions. Then I will discuss a few issues fundamental to such an endeavor. 1.1. A Theory of the Mind and Beyond Before embarking on this journey it might help to make clear at the outset that what is to be described and discussed in the present book including concepts theories models and simulations is centered on a particular theoretical framework—namely the Clarion framework. It is worth noting that Clarion in its full-fedged form is a generic and rela - tively comprehensive theory of the human mind 1 along with a computa- tional implementation of the theory. It is thus a computational “cognitive 1. “Mind” is a complex notion. Rather than engaging in a philosophical discourse on the notion the focus here is instead on mechanisms and processes of the mind. In turn “mechanism” here refers to physical entities and structures and their properties that give rise to certain characteristics of the mind. Although living things often appear to have certain characteristics that have no counterpart in the physical universe one may aim to go beyond these appearances Thagard 1996.

slide 21:

2 Chapter 1 architecture” as is commonly referred to in cognitive science cognitive psychology or more generally in the “cognitive sciences”. 2 In general a cognitive architecture is a broad domain-generic cognitive-psychological model implemented computationally. Clarion has been in continuous development for a long time at least since 1994 although its predecessors have had a longer history. It has been aimed to capture explain and simulate a wide variety of cogni- tive-psychological phenomena within its unifed framework thus leading hopefully and ultimately to unifed explanations of psychological and even other related phenomena as advocated by e.g. Newell 1990. The exact extent of cognitive-psychological phenomena that have been captured and explained within its framework will be discussed in detail in subsequent chapters. It is not unreasonable to say that Clarion constitutes a relatively comprehensive theory of the mind or at least an initial ver- sion of such a theory. In fact Clarion within itself contains several different kinds of theo- ries. First it contains a core theory of the mind at a conceptual level. It posits essential theoretical distinctions such as implicit versus explicit pro- cesses action-centered versus non-action-centered processes and so on as well as their relationships Sun 2002 2012. With these distinctions and other high-level constructs it specifes a core theory of the essential structures mechanisms and processes of the mind at an abstract concep- tual level Sun Coward and Zenzen 2005. Second it also contains a more detailed but still generic compu- tational model implementing the abstract theory. This implementation constitutes what is usually referred to as a computational cognitive archi- tecture: that is a generic computational cognitive i.e. psychological model describing the architecture of the mind which by itself also con- stitutes a theory of the mind albeit at a more detailed and computational level as will be argued later see also Sun 2009b. 2. In the narrow sense “cognition” refers to memory learning concepts decision making and so on—those aspects of the individual mind that are not directly related to motivation emotion and the like. In the broadest sense it may refer to all aspects of the individual mind especially when methods and perspectives from contemporary cogni- tive science are used in studying these aspects. In the latter case I often use a hyphen- ated form “cognition-psychology” to make it clear. However the plural form “cognitive sciences” is often used to refer to all felds of cognitive behavioral and psychological sciences applying the broadest sense of the term. Similarly in the term “cognitive archi- tecture” the word “cognitive” should be interpreted in the broadest sense.

slide 22:

What Is A Cognitive Architecture 3 Third with the generic computational cognitive architecture one may construct specifc models and simulations of specifc psychological phe - nomena or processes. That is one may “derive” specifc computational models namely specifc computational theories for specifc psycho - logical phenomena or processes from the generic computational model theory. So the generic theory leads to specifc theories. Clarion encompasses all of the above simultaneously. Thus it syn- thesizes different types of theories at different levels of theoretical abstraction Sun 2009b. Below I will refer alternately to Clarion in these different senses at different levels of abstraction as appropriate. 1.2. Why Computational Models/Theories Why would one want computational models for the sake of under- standing the human mind Why are computational models useful exactly Generally speaking models of various forms and complexities may be roughly categorized into computational mathematical and verbal-conceptual varieties Sun 2008. Computational models present algorithmic descriptions of phenomena often in terms of mechanistic and process details. Mathematical models present often abstract relation- ships between variables using mathematical equations. Verbal-conceptual models describe entities relations or processes in informal natural lan- guages such as English. A model regardless of its genre might often be viewed as a theory of whatever phenomena that it purports to capture. This point has been argued extensively before by e.g. Newell 1990 and Sun 2009b. Although each of these types of models has its role to play I am mainly interested in computational modeling. The reason for this preference is that at least at present computational modeling appears more promising in many respects. It offers the expressive power that no other approach can match because it provides a wider variety of modeling techniques and methodologies. In this regard note that mathematical models may be viewed as a subset of computational models because normally they can lead readily to computational implementations even though some of them may be sketchy not covering suffcient mechanistic or process details. Computational modeling also supports practical applications see e.g. Pew and Mavor 1998 Sun 2008.

slide 23:

4 Chapter 1 Computational models are mostly mechanistic and process oriented. That is they are mostly aimed at answering the questions of how human performance comes about by what psychological structures mechanisms and processes and in what ways. 3 The key to understanding cognitive- psychological phenomena is often in fne details which computational modeling can describe and illuminate Newell 1990 Sun 2009b. Computational models provide algorithmic specifcity: detailed exactly specifed and carefully worked-out steps arranged in precise and yet fex - ible sequences. Thus they provide clarity and precision see e.g. Sun 2008. Computational modeling enables and in fact often forces one to think in terms of mechanistic and process details. Instead of verbal-conceptual theories which may often be vague one has to think clearly algorith- mically and in detail when dealing with computational models/theo- ries. Computational models are therefore useful tools. With such tools researchers must specify a psychological mechanism or process in suff - cient detail to allow the resulting models to be implemented on comput- ers and run as simulations. This requires that all elements of a model e.g. all its entities relationships and so on be specifed exactly. Thus it leads to clearer more consistent more mechanistic more process-oriented the- ories. Richard Feynman once put it this way: “What I cannot create I do not understand.” This applies to the study of human cognition-psychol- ogy. To understand is to create in this case on a computer at least. Computational models may be necessary for understanding a system as complex and as internally diverse as the human mind. Pure mathemat- ics developed mainly for describing the physical universe may not be suffcient for understanding a system as different as the human mind. Compared with theories developed in other disciplines such as phys- ics computational modeling of the mind may be mathematically less “elegant” but the human mind itself may be inherently less mathemati- cally elegant when compared with the physical universe as argued by e.g. Minsky 1985. Therefore an alternative form of theorizing may be necessary—a form that is more complex more diverse and more algorithmic in nature. Computational modeling provides a viable way 3. It is also possible to formulate so called “product theories” which provide a func- tional account of phenomena but do not commit to a particular psychological mechanism or process. Thus product theories can be evaluated mainly by product measures. One may also term product theories black-box theories or input-output theories.

slide 24:

What Is A Cognitive Architecture 5 of specifying complex and detailed theories of cognition-psychology. Therefore they may be able to provide unique explanations and insights that other experimental or theoretical approaches cannot easily provide. A description or an explanation in terms of computation that is per- formed in the mind/brain can serve either as a fne-grained specifcation of cognitive-psychological processes underlying behavior roughly the mind or as an abstraction of neurobiological and neurophysiological data and discoveries roughly the brain among other possibilities that may also exist. In general it is not diffcult to appreciate the usefulness of a computational model in this regard in either sense especially one that summarizes a body of data which has been much needed in psychology and in neuroscience given the rapid growth of empirical data. In particular understanding the mind at the psychological level through computational modeling may be very important. One would naturally like to know more about both the mind and the brain. So far at least we still know little about the biology and physiology of the brain relatively speaking. So for this reason and others we need a higher level of abstraction that is we need to study the mind at the psychological level in order to make progress toward the ultimate goal of fully under- standing the mind and the brain. Trying to fully understand the human mind purely from observations of human behavior e.g. strictly through behavioral experiments is likely untenable except perhaps for small limited task domains. The rise and fall of behaviorism is a case in point. This point may also be argued on the basis of analogy with the physical sciences as was done in Sun Coward and Zenzen 2005. The processes and mechanisms of the mind cannot be understood purely on the basis of behavioral experiments which often amount to tests that probe relatively superfcial features of human behav - ior further obscured by individual and cultural differences and other con- textual factors. It would be extremely hard to understand the human mind in this way just like it would be extremely hard to understand a complex computer system purely on the basis of testing its behavior if one does not have any prior ideas about the inner workings and theoreti- cal underpinnings of that system Sun 2007 2008 2009b. Experimental neuroscience alone may not be suffcient either at least for the time being. Although much data has been gathered from empirical work in neuroscience there is still a long way to go before all the details of the brain are identifed let alone the psychological functioning on that basis. Therefore as argued earlier at least at present it is important to

slide 25:

6 Chapter 1 understand the mind/brain at a higher level of abstraction. Moreover even when we fnally get to know all the minute details of the brain we would still need a higher-level yet precise mechanistic process-based understanding of its functioning. Therefore we still need a higher level of theorizing. In an analogous way the advent of quantum mechanics did not eliminate the need for classical mechanics. The progress of chemistry was helped by the discoveries in physics but chemistry was not replaced by physics. It is imperative that we also investigate the mind at a higher level of abstraction beyond neuroscience. Computational modeling has its unique indispensable and long-term role to play especially for gaining conceptually clear detailed and principled understanding of the mind/ brain. It might be worth mentioning that there have been various view- points concerning the theoretical status of computational modeling. For example many believed that a computational model and computational simulation on its basis may serve as a generator of phenomena and data. That is they are useful media for hypothesis generation. In particular one may use simulation to explore process details of a psychological phe- nomenon. Thus a model is useful for developing theories constituting a theory-building tool. A related view is that computational modeling and simulation are suitable for facilitating a precise instantiation of a preex- isting verbal-conceptual theory e.g. through exploring possible details for instantiating the theory and consequently detailed evaluations of the theory against data. These views however are not incompatible with a more radical position e.g. Newell 1990 Sun 2009b that a computa- tional model may constitute a theory by itself. It is not the case that a model is limited to being built on top of an existing theory being applied for the sake of generating data being applied for the sake of validating an existing theory or being used for the sake of building a future theory. According to this more radical view a model may be viewed as a theory by itself. In turn algorithmic descriptions of computational models may be considered just another language for specifying theories Sun 2009b Sun 2008. 4 The reader is referred to Sun 2009b for a more in-depth discussion of this position. 4. Constructive empiricism van Fraasen 1980 may serve as a philosophical founda- tion for computational cognitive modeling compatible with the view of computational models as theories Sun 2009b.

slide 26:

What Is A Cognitive Architecture 7 In summary computational models theories can be highly useful to psychology and cognitive science when viewed in the light above and when the issues discussed below are properly addressed. 1.3. Questions about Computational Models/Theories There are of course many questions that one can and should ask about any computational model before “adopting” it in any way. One important question about any particular computational model is this: how much light can it really shed on the phenomena being mod- eled There are a number of aspects to this question for instance: • Do the explanations provided by the computational model capture accurately human “performance” in a Chomskian sense Chomsky 1980 That is does it capture and explain suffciently the subtleties exhibited in the empirical data If an explanation is devoid of “performance” details as observed in empirical data it will be hard to justify the appropriateness of such an explanation especially when there are other possible ways of describing the data. 5 • Does the model take into consideration higher-level or lower-level constraints above or below the level of the model in question There are usually many possible models/theories regarding some limited data. Higher-level or lower-level consid- erations among other things may be used to narrow down the choices. • Does the model capture in a detailed way psychological mechanisms and processes underlying the data If a model lacks mechanistic process-oriented details it may be less likely to bring new insights into explaining the dynamics underlying the data. • Do the primitives entities structures and operations used in the model provide some descriptive power and other advan- tages over and above other possible ways of describing human behavior and performance but without being overly generic 5. This is not the case for Noam Chomsky’s theory of language which thus serves as an exception.

slide 27:

8 Chapter 1 • Does the model provide a basis for tackling a wide set of cognitive-psychological tasks and data If a model is insuffcient in terms of breadth of coverage it cannot claim to be a “general” theory. It should be noted that in relation to the issue of generality one should be aware of the danger of over-generality. That is a model might be so under-constrained that it may match practically any possible data real- istic or unrealistic. To address this problem many simulations in a wide range of domains are needed in order to narrow down choices and to constrain parameter spaces more on this in Chapter 8. From the point of view of the traditional cognitive science a model/ theory at the computational or knowledge level in Marr’s 1982 or Newell’s 1990 sense can provide a formal language for describing a range of cognitive-psychological tasks. Indeed in the history of cogni- tive science some high-level formal theories were highly relevant e.g. Chomsky’s theory of syntactic structures of language. So a further question is: • How appropriate is the model/theory in terms of providing a “formal language” for a broad class of tasks or data Does it have realistic expressiveness suffcient for the target tasks or data but not much more or less and realistic constraints of various types at various levels Furthermore what is more important than a formal e.g. mathemati- cal or computational language for describing cognition-psychology is the understanding of the “architecture” of the mind especially in a mecha- nistic computational sense. That is one needs to address the following question: • How do different components of the mind interact and how do they ft together Correspondingly how do different com - ponents of a computational model/theory interact and how do these different components ft together instead of just a mere collection of limited models Studying architectural issues may help us to gain new insight narrow down possibilities and constrain the components involved. Moreover different components and different functionalities of the mind for example perception categorization concepts memory

slide 28:

What Is A Cognitive Architecture 9 decision making reasoning problem solving planning communication action learning metacognition and motivation all interact with and depend on each other. Their patterns of interaction change with chang- ing task demands growing personal experiences varying sociocultural contexts and milieus and so on. Some argue that cognition-psychology represents a context-sensitive dynamic statistical structure that on the surface at least changes constantly—a structure in perpetual motion. However complex dynamic systems may be attributed to its constituting elements. Thus one may strive for a model that captures the dynamics of cognition-psychology through capturing its constituting elements and their interaction and dependency. So an important question is: • How does a model/theory account for the dynamic nature of cognition-psychology Finally one has to consider the cost and beneft of computational modeling: • Is the complexity of a model/theory justifed by its explanatory utility considering all the questions above These questions cannot be addressed in abstraction. My specifc answers to them in the context of Clarion will emerge in subsequent chapters as details of Clarion emerge in these chapters. 1.4. Why a Computational Cognitive Architecture Among different types of computational cognitive-psychological models/ theories computational cognitive architectures stand out. A computa- tional cognitive architecture as commonly termed in cognitive science is a broadly scoped domain-generic cognitive-psychological model imple- mented computationally capturing the essential structures mechanisms and processes of the mind to be used for broad multiple-level multi- ple-domain analysis of behavior e.g. through its instantiation into more detailed computational models or as a general framework Newell 1990 Sun 2007. Let us explore this notion of cognitive architecture with a comparison. The architecture for a building consists of its overall structural design and major constituting structural elements such as external walls foors roofs

slide 29:

10 Chapter 1 stairwells elevator shafts and so on. Furniture can be easily rearranged or replaced and therefore may not be part of the architecture. By the same token a cognitive architecture includes overall structures essential divisions of modules e.g. subsystems essential relations between mod- ules basic representations and algorithms within modules and a variety of other major aspects Sun 2007 Langley Laird Rogers 2009. In general a cognitive architecture includes those aspects that are relatively invariant across time domains and individuals. It deals with them in a structurally and mechanistically well-defned way. A cognitive architecture can be important to understanding the human mind. It provides concrete computational scaffolding for more detailed modeling and exploration of cognitive-psychological phenomena and data through specifying essential computational structures mechanisms and processes. That is it facilitates more detailed modeling and explora- tion of the mind. As discussed earlier computational cognitive modeling explores cognition-psychology through specifying computational mod- els of cognitive-psychological mechanisms and processes. It embodies descriptions of cognition-psychology in computer algorithms and pro- gram codes thereby producing runnable models. Detailed simulations can then be conducted. In this undertaking a cognitive architecture can be used as the unifying basis for a wide range of modeling and simulation. Note that here I am mainly referring to psychologically oriented cogni- tive architectures as opposed to software engineering oriented cognitive architectures which are quite different in terms of purpose. A cognitive architecture serves as an initial set of relatively generic assumptions that may be applied in further modeling and simulation. These assumptions in reality may be based on empirical data philosoph- ical arguments or computational considerations. A cognitive architecture is useful and important because it provides a relatively comprehensive yet precise foundation that facilitates further modeling in a wide variety of domains Cooper 2007. In exploring cognitive-psychological phenomena the use of cog- nitive architectures forces one to think in terms of mechanistic and process-oriented details. Instead of using often vague and underspecifed verbal-conceptual theories cognitive architectures force one to think more clearly. Anyone who uses cognitive architectures must specify a cognitive-psychological mechanism or process in suffcient detail to allow the resulting models to run as simulations. This approach encour- ages more detailed and clearer theories. It is true that more specialized

slide 30:

What Is A Cognitive Architecture 11 narrowly scoped computational models may also serve this purpose but they are not as generic and as comprehensive. Consequently they are not as generally useful. Cognitive architectures are thus crucial tools Pew and Mavor 1998 Sun 2007. A cognitive architecture may also provide a deeper level of explana- tion Sun 2007. Instead of a model specifcally designed for a specifc task which is often ad hoc a cognitive architecture naturally encourages one to think in terms of the mechanisms and processes available within a generic model that are not specifcally designed for a particular task and thereby to generate explanations of the task that are not centered on superfcial high-level features of the task as often happens with spe - cialized narrowly scoped models—that is to generate explanations of a deeper kind. To describe a task in terms of available mechanisms and processes of a cognitive architecture is to generate explanations based on primitives of cognition-psychology envisioned in the cognitive architec- ture thereby leading to deeper explanations. Because of the nature of such deeper explanations this approach is also more likely to lead to unifed explanations for a wide variety of data and phenomena because potentially a wide variety of tasks data and phenomena can be explained on the basis of the same set of primitives provided by the same cognitive architecture Sun 2007. Therefore a cognitive architecture is more likely to lead to a unifed comprehensive theory of the mind unlike using more specialized narrowly scoped mod- els Newell 1990. Although the importance of being able to reproduce the nuances of empirical data is evident broad functionalities in cognitive architectures are even more important Newell 1990. The human mind needs to deal with all of the necessary functionalities: perception categorization memory decision making reasoning planning problem solving commu- nication action learning metacognition motivation and so on. The need to emphasize generic models capable of broad functionalities arises also because of the need to avoid the myopia often resulting from narrowly scoped research. For all of these reasons above developing cognitive architectures is an important endeavor in cognitive science. It is of essential importance in advancing the understanding of the human mind Sun 2002 2004 2007. Existing cognitive architectures that have served this purpose include ACT-R Soar Clarion and a number of others see e.g. Taatgen and Anderson 2008 for a review.

slide 31:

12 Chapter 1 In addition cognitive architectures also in a way support the goal of general AI that is building artifcial systems that are as capable as human beings. In relation to building intelligent systems a cognitive architecture may provide the underlying infrastructure because it may include a vari- ety of capabilities modules and subsystems that an intelligent system needs. On that basis application systems may be more readily developed. A cognitive architecture carries with it theories of cognition-psychology and understanding of intelligence gained from studying the human mind. In a way cognitive architectures reverse engineer the only truly intel- ligent system around—the human mind. Therefore the development of intelligent systems on that basis may be more cognitively-psychologically grounded which may be advantageous in some circumstances. The use of cognitive architectures in building intelligent systems may also facili- tate the interaction between humans and artifcially intelligent systems because of the relative similarity between humans and cognitively-psy- chologically based intelligent systems. It was predicted a long time ago that “in not too many years human brains and computing machines will be coupled together very tightly and the resulting partnership will think as no human brain has ever thought …” Licklider 1960. Before that hap- pens a better understanding of the human mind is needed especially a better understanding in a computational form. There are of course questions that one should ask about cognitive architectures in addition to or instantiating questions about computa- tional modeling in general as discussed earlier. For instance a cognitive architecture is supposed to include all essential psychological capabili- ties and functionalities. As mentioned before those functionalities may include perception categorization memory decision making reasoning problem solving communication action and learning. They may involve all kinds of representation in a broad sense. There are also motivational and metacognitive processes. However currently most cognitive archi- tectures do not yet support all of these functionalities at least not fully. So what is minimally necessary How should these functionalities inter- act To what extent are they separate And so on. There are no simple answers to these questions but they will be addressed along the way in this book. In this regard a question concerning any capability in a cognitive architecture is whether the cognitive architecture includes that capabil- ity as an integral part or whether it includes suffcient basic functional - ities that allow the capability to emerge or to be implemented later on.

slide 32:

What Is A Cognitive Architecture 13 This may be determined by what one views as an integral part of a cogni- tive architecture and what one views as a secondary or derived capabil- ity. Sun 2004 provides a discussion of the relation between a cognitive architecture and the innate structures in the human mind and the notion of minimality in a cognitive architecture. These ideas may help to sort out what should or needs to be included in a cognitive architecture Sun 2004. The outcomes of the deliberation on this and other questions will be presented in the subsequent chapters. 1.5. Why Clarion Among existing cognitive architectures why should one choose Clarion In a nutshell one might prefer Clarion for the totality of the following reasons: • Clarion is a cognitive architecture that is more comprehensive in scope than most other cognitive architectures in existence today as will become clear later. • Clarion is psychologically realistic to the extent that it has been validated through simulating and explaining a wide variety of psychological tasks data and phenomena as detailed in chapters 5 6 and 7. • Its basic principles and assumptions have been extensively argued for and justifed in relation to a variety of different types of evidence as detailed in chapters 2 3 and 4. • It has major theoretical implications as well as some practical relevance. It has provided useful explanations for a variety of empirical data leading to a number of signifcant new theories regarding psychological phenomena e.g. Sun Slusarz Terry 2005 Helie Sun 2010. • In addition to addressing problems at the psychological level it has also taken into account higher levels for example regarding social processes and phenomena as well as lower levels Sun Coward Zenzen 2005. More specifcally Clarion has been successful in computationally mod - eling simulating accounting for and explaining a wide variety of psy- chological data and phenomena. For instance a number of well-known

slide 33:

14 Chapter 1 skill-learning tasks have been simulated using Clarion that span the entire spectrum ranging from simple reactive skills to complex cognitive skills. The simulated tasks for example include serial reaction time tasks arti- fcial grammar learning tasks dynamic process control tasks alphabetical arithmetic tasks and Tower of Hanoi e.g. Sun Slusarz Terry 2005 Sun 2002. In addition extensive work has been done in modeling com- plex and realistic skill-learning tasks that involve complex sequential deci- sion making Sun et al. 2001. Furthermore many other kinds of tasks not usually dealt with by cognitive architectures—reasoning tasks social simulation tasks as well as metacognitive and motivational tasks—have been tackled by Clarion. While accounting for various psychological tasks data and phenomena Clarion provides explanations that shed new light on underlying cognitive-psychological processes. See for example Sun et al. 2001 Sun Slusarz and Terry 2005 Sun Zhang and Mathews 2006 and Helie and Sun 2010 for various examples of such simula- tions and explanations. These simulations more importantly provided insight that led to some major new theories concerning a number of important psychological functionalities. Some new theories resulting from Clarion include: • The theory of bottom-up learning from implicit to explicit learning as developed in Sun et al. 2001. • The theory of the implicit-explicit interaction and their syner- gistic effects on skill learning as developed in Sun Slusarz and Terry 2005. • The theory of creative problem solving as described in Helie and Sun 2010. • The theory of human motivation and its interaction with cogni- tion as described in Sun 2009 as well as in related simulation papers e.g. Wilson Sun Mathews 2009 Sun Wilson 2011 Sun Wilson 2014 • The theory of human reasoning based on implicit and explicit representation and their interaction as developed in Sun 1994 1995 and Sun and Zhang 2006. These theories are standalone conceptual-level theories of psychological phenomena. However these theories are also an integral part of Clarion. They have been computationally instantiated. They have led not only to

slide 34:

What Is A Cognitive Architecture 15 numerical quantitative simulations but also to major qualitative theo- retical predictions. I should mention here that two meta-principles have guided the devel- opment of this cognitive architecture: a completeness of functional- ities to include as many functionalities as possible but b parsimony of mechanisms to reduce the number of distinct mechanisms and their complexity as much as possible. Or to put it another way the goal for Clarion has been: maximum scope and minimum mechanism. That goal and the associated meta-principles have led to the aforementioned theo- ries and explanations by Clarion. Given all of the above Clarion is worthy of further exploration and examination. In particular its comprehensive scope should be examined in more detail. Thus a book-length treatment is required. 1.6. Why This Book Although a substantial number of articles including journal and confer- ence papers have been published on Clarion and its modeling of psycho- logical data of various kinds there is currently no one single volume that contains all of the information especially not in a unifed and accessible form. Therefore it seems a good idea to put together a single volume for the purpose of cataloguing and explaining in a unifed and accessible way what has been done with regard to Clarion and why it might be of interest. Furthermore a book may contain much more material than a typical journal or conference paper. It may describe not only details of Clarion but also many detailed models of psychological phenomena based on Clarion. It may summarize materials published previously in addition to new materials. A book may also provide theoretical and meta-theoretical discussions of issues involved. Above all a book may provide a gentler introduction to Clarion and its exploration of psychological mechanisms and processes which may be of use to some readers. The present book will present a unifed albeit preliminary and still incomplete view of the human mind and interpret and explain empirical data on the basis of that view. The focus will be on broad interpretations of empirical data and phenomena emphasizing unifed explanations of a wide variety of psychological tasks and data. Thus

slide 35:

16 Chapter 1 exact parameter values and other minute technical details will be minimized. For the sake of clarity I will proceed in a hierarchical fashion. In other words there will be a series of progressively more detailed descrip- tions. First a high-level conceptual sketch will be given then a more detailed description will be provided. After that there will be an even more detailed more technical description. However the most techni- cally exact and complete description with a code library can be found in a forthcoming companion technical book on Clarion. In this way the reader may stop at any time up to the level where he or she feels comfortable. I will start with the overall Clarion framework and then move on to individual components or aspects. To achieve clarity I will limit the amount of details discussed to only those that are minimally necessary. Fortunately the technical book will provide full technical specifca - tions. With regard to technical details especially in relation to simu- lations I will have to strike a balance between conceptual clarity and technical specifcity. Of course both are important. To achieve concep - tual clarity a high-level conceptual explanation will be provided. To achieve some technical specifcity a more technical computational description or explanation will also be provided corresponding to the high-level conceptual explanation. 1.7. A Few Fundamental Issues To start I will quickly sketch a few foundational issues. My stands on these issues form the meta-theoretical basis of Clarion. Details of the cognitive architecture will be explained in subsequent chapters. 1.7.1. Ecological-Functional Perspective The development of a cognitive architecture needs to take into con- sideration of what I have called the ecological-functional perspective. As discussed in Sun 2012 and Sun 2002 the ecological-functional perspective includes a number of important considerations on human cognition-psychology especially in relation to ecological realism of

slide 36:

What Is A Cognitive Architecture 17 cognitive-psychological theories or models. They may be expressed as dictums such as: • taking into account ecological niches evolutionarily or at the present and focusing attention on characteristics of everyday activities that are most representative of the ecological niches Sun 2002 more later • taking into account the role of function because cognitive-psychological characteristics are often if not always functional useful in some way for everyday activities within an ecological niche • taking into account cost-beneft trade-offs of cognitive-psychological characteristics such as implicit versus explicit processes 6 as psychological characteristics are often selected based on cost-beneft considerations evolutionarily or at the present. In particular these dictums imply that human cognition-psychology is mostly activity-based action-oriented and embedded in the world. They also seem to point toward implicit subconscious or unconscious psychological processes as opposed to focusing exclusively on explicit processes. Humans often interact with the world in a rather direct and unmediated way Heidegger 1927 Dreyfus 1992 Sun 2002. These dictums serving as meta-heuristics for developing cognitive architectures will become clearer in the next chapter when the justifca - tions for the essential framework of Clarion are discussed. 1.7.2. Modularity Fodor 1983 argued that the brain/mind was modular and its modules worked largely independently and communicated only in a limited way. However evidence to the contrary has accumulated that modules and subsystems in the brain/mind may instead be more richly interconnected anatomically and functionally Damasio 1994 Bechtel 2003. Nevertheless starting off with a modular organization might make the task of understanding the architecture of the human mind more tractable. 6. For instance compared with implicit processes explicit processes may be more precise but may be more effortful. See more discussions in Chapter 3.

slide 37:

18 Chapter 1 Connections communications and interactions if necessary may be added subsequently. At a minimum some cognitive-psychological func- tionalities do appear to be specialized and somewhat separate from others in a sense. They may be so either because they are functionally encapsu- lated their knowledge mechanisms and processes do not transfer easily into other domains or because they are physically neurophysiologi- cally encapsulated. Modularity can be useful functionally for example to guarantee effciency or accuracy of important or critical behaviors and routines whether they are a priori or learned or to facilitate parallel operations of multiple processes Sun 2004. Hence we start with a cir- cumscribed modular view. 1.7.3. Multiplicity of Representation With modularity i.e. with the co-existence of multiple modules mul- tiple different representations either in terms of form or in terms of content may co-exist. Here I use the term “representation” to denote any form of internal encoding either explicitly and individually encoded or embodied/ enmeshed within a complex mechanism or process. Thus this notion of “representation” is not limited to explicit individuated symbolic entities and their structures as often meant by “representationalism”. Because it is not limited to symbolic forms it includes for example connectionist encoding dynamic system content and so on. So the term should be inter- preted broadly here. In terms of representational form there are for example symbolic- localist representation and distributed connectionist representation. Symbolic-localist representation implies representing each unique con- cept by a unique basic representational entity such as a node in a net- work. Distributed representation involves representing each concept by an activation pattern over a shared set of nodes in a network Rumelhart et al. 1986. Different forms of representations have different computa- tional characteristics: for example crisp versus graded rule-based versus similarity-based one-shot learning versus incremental learning and so on as will be discussed in more detail later. In terms of representational content there may be the following types: procedural representation declarative representation metacognitive representation motivational representation and so on. Each of these types

slide 38:

What Is A Cognitive Architecture 19 is necessary for a full account of the human mind. In subsequent chapters when I discuss each of these types in turn I will present arguments why each of them is needed. Each type may in turn involve multiple represen- tational forms within. On the other hand one may question why an individual needs multiple representational forms after all. There are a number of potential advan- tages that may be gained by involving multiple representational forms. For example in incorporating both symbolic-localist and distributed represen- tation for capturing explicit and implicit knowledge respectively as will be detailed later one may gain • synergy in skill learning from dual procedural representation • synergy in skill performance from dual procedural representation • synergy in reasoning from dual declarative representation and so on. These advantages have been demonstrated before in previous publications I will elaborate on these advantages in later chapters when I revisit these points. 1.7.4. Dynamic Interaction In a cognitive architecture various modules in the previously dis- cussed sense have to work with each other to accomplish psychologi- cal functioning. Modules of different kinds and sizes e.g. subsystems and components within each subsystem interact with each other dynamically. At the highest level the interaction among subsystems may include metacognitive monitoring and regulation of other processes i.e. the interaction between the metacognitive subsystem and the other subsys- tems. The interaction among subsystems may also involve motivated action decision making i.e. the interaction between the motivational subsystem and the action-centered subsystem. Within each subsystem many component modules exist and they also interact with each other necessary for accomplishing cognitive-psychological functioning. Note that these characteristics may not have been suffciently captured by most existing cognitive-psychological models includ- ing cognitive architectures. Compared with these other models Clarion is unique in terms of containing well-developed built-in motivational constructs and well-developed built-in metacognitive

slide 39:

20 Chapter 1 constructs. These are not commonly found in existing cognitive archi- tectures. Nevertheless I believe that these features are crucial to a cog- nitive architecture because they capture important or indispensable elements of the human mind necessary in the interaction between an individual and his or her physical and social world Sun 2009. Details will be presented in subsequent chapters. 1.8. Concluding Remarks So far I have covered only some preliminary ideas which are necessary background regarding cognitive architectures. The questions that have been addressed include: Why should one use computational modeling for studying cognition-psychology Why should one use cognitive archi- tectures among other computational models Why should one use the Clarion cognitive architecture among other possible cognitive architec- tures And other questions and issues. More importantly the basic “philosophy” in regard to a number of fun- damental issues has been outlined. In particular the principles of modu- larity multiplicity of representation and dynamic interaction include that among motivation cognition and metacognition are of fundamen- tal importance to Clarion. The rest of the book is divided into eight chapters. They include three chapters for presenting various theoretical conceptual and technical aspects of Clarion three chapters on various simulations using Clarion and additional materials in the remaining two chapters. Finally a note for the interested reader: for general surveys discussions and comparisons of computational cognitive architectures in the context of cognitive-psychological modeling covering other well-known cogni- tive architectures such as ACT-R and Soar see Pew and Mavor 1998 Ritter et al. 2003 Sun 2006 Chong Tan and Ng 2007 Taatgen and Anderson 2008 Langley et al. 2009 Thórisson and Helgasson 2012 Helie and Sun 2014b among other existing publications see also Chapter 9.

authorStream Live Help