The Cognitive Psychology of Planning (Current Issues in Thinking & Reasoning)

THE COGNITIVE PSYCHOLOGY OF PLANNING Current Issues in Thinking and Reasoning Published Titles Working Memory and Thi...

Author: Robin Morris

104 downloads 1158 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

THE COGNITIVE PSYCHOLOGY OF PLANNING

Current Issues in Thinking and Reasoning Published Titles Working Memory and Thinking (Robert H. Logie and Kenneth J. Gilhooly) 1998 Imagery, Language and Visuo-Spatial Thinking (Michel Denis, Robert H. Logie, Cesare Cornoldi, Manuel de Vega, Johannes Engelkamp) 2001 Evolution and the Psychology of Thinking: The Debate (Ed. David E. Over) 2003 Methods of Thought: Individual Differences in Reasoning Strategies (Maxwell Roberts and Elizabeth Newton) in production

The Cognitive Psychology of Planning

Edited by

Robin Morris and Geoff Ward

First published 2005 by Psychology Press 27 Church Road, Hove, East Sussex BN3 2FA Simultaneously published in the USA and Canada by Psychology Press 270 Madison Avenue, New York, NY 10016

This edition published in the Taylor & Francis e-Library, 2004. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Psychology Press is a part of T&F Informa plc Copyright © 2005 Psychology Press All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. The publisher makes no representation, express or implied, with regard to the accuracy of the information contained in this book and cannot accept any legal responsibility or liability for any errors or omissions that may be made. This publication has been produced with paper manufactured to strict environmental standards and with pulp derived from sustainable forests. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data The cognitive psychology of planning / [edited by] Robin Morris and Geoff Ward.— 1st ed. p. cm.— (Current issues in thinking & reasoning) Includes bibliographical references and index. ISBN 1-84169-333-2 (hardcover) 1. Planning—Psychological aspects. 2. Cognitive psychology. I. Morris, Robin, 1958– II. Ward Geoff, 1968– III. Series BF433.P6C64 2004 153.4—dc22 2004011671

ISBN 0-203-49356-7 Master e-book ISBN

ISBN 1-84169-333-2 (Print Edition)

Contents

List of contributors

ix

1. Introduction to the psychology of planning Geoff Ward and Robin Morris Introduction 1 Theoretical background 2 Methodology and planning 24 Neuropsychology and planning 26 Overview of the chapters in this volume References 32

1

27

2. Planning and problem solving in well-defined domains Simon P. Davies Introduction 35 Characterizing planning behaviour in well-defined domains 36 Selection and effectiveness of different planning behaviours 39 Problem complexity and planning strategy 39 Problem-solving environment and planning strategy Individual and group differences and planning strategy 44 Effectiveness of initial planning 46 Summary and conclusions 48 References 48

35

41

v

vi

CONTENTS

3. Planning and ill-defined problems 53 Thomas C. Ormerod Introduction 53 Planning in puzzle solving 55 Plans, planning and expert skill 60 Conclusions 67 References 69 4. Working memory and planning 71 K. J. Gilhooly Introduction 71 General strategies for developing plans of action 73 Working memory in computational models of problem solving and planning 75 Multi-component approaches to working memory 76 Single resource approaches to working memory 78 “Move” tasks: The Tower of London (TOL) 78 Studies of planning and working memory in chess 83 Concluding comments 85 References 85 5. Planning and the executive control of thought and action 89 Geoff Ward Introduction 89 Do we possess “lower level” action plans? 92 Lower order planning: How do we control the initiation of lower level action plans? 93 “Higher order” planning I: Planning what to do to solve a problem 96 “Higher order” planning II: Planning when to do things to solve a problem 99 When and why do we plan? 105 Summary 107 References 108 6. Adult ageing and cognitive planning 111 Louise H. Phillips, Mairi S. MacLeod, and Matthias Kliegel Age, the frontal lobes and executive function 111 Effects of ageing on the Tower of London (TOL) planning task 113 Formulating complex plans in the laboratory 115 Age and action planning: Six Elements Task (SET) 119 Age and errand planning in a naturalistic setting 122 Comparison of age effects on realistic and abstract planning tasks Adult ageing and planning: Themes emerging from the literature 125 References 131

124

CONTENTS

vii

7. Cognitive planning in Humans: New insights from the Tower of London (TOL) task 135 Adrian M. Owen Introduction 135 The Tower of London (TOL) 136 Psychological considerations: Cognitive components of performance identified from eye-tracking behaviour 139 Anatomical considerations: Localizing the core neural substrates of performance using functional neuroimaging 142 Further considerations: Is planning just working memory for the future? 146 Conclusions 148 References 148 8. Planning in patients with focal brain damage: From simple to complex task performance 153 Robin Morris, Maria Kotitsa, and Jessica Bramham Introduction 153 Illustrative case examples 154 Development of strategies 156 Problem solving on the Tower of Hanoi (TOH) 158 Planning and organizational abilities investigated through simulation Virtual reality exploration of strategy formation, rule breaks and prospective memory 170 Conclusions 177 References 177 9. Planning and the brain 181 Jordan Grafman, Lee Spector, and Mary Jo Rattermann Introduction 181 Cognitive and computational perspectives on planning Cognitive neuroscience perspectives 185 Conclusions 191 References 195

182

10. The search for specific planning processes 199 Paul Burgess, Jon Simons, Laure Coates, and Shelley Channon Introduction 199 Challenge 1: Is “planning” just a label for a range of disparate human activities? 200 Challenge 2: Planning as the expression of stored preferences 202 Challenge 3: Is the existing experimental evidence consistent with the assumption that “look-ahead” is the principal construct underpinning planning performance? 204 Summary so far 210 Experimental evidence from our laboratory 211

166

viii

CONTENTS

Planning using real-life analogue tasks 213 Evidence from neurological patients 217 Planning deficits and localization 219 Planning and “construct validity” 220 Conclusion 221 References 224 Author index Subject index

229 239

List of contributors

Jessica Bramham, Department of Psychology, Institute of Psychiatry, De Crespigny Park, London SE5 8AF, UK Paul Burgess, Institute of Cognitive Neuroscience & Dept of Psychology, University College London, Alexandra House, 17 Queen Square, London WC1N 3AR, UK Shelley Channon, Department of Psychology, University College London, Gower Street, London WC1N 3AR, UK Laure Coates, Department of Clinical Psychology, Institute of Psychiatry, King’s College Healthcare Trust, Denmark Hill, London SE5 8AZ, UK Simon P. Davies, Department of Psychology, University of Hull, Hull HU6 7RX, UK Ken J. Gilhooly, School of Psychology, University of Hertfordshire, Hatfield ALD 9AB, UK Jordan Grafman, Chief, Cognitive Neuroscience Section, National Institute of Neurological Disorders and Stroke, Building 10, Room 5C205, 10 Center Drive, MSC 1440, Bethesda, Maryland 20892-1440, USA Matthias Kliegel, Assistant Professor, Department of Gerontopsychology, Institute for Psychology, University of Zurich, Schaffhauserstr. 15, Ch-8006 Zurich, Switzerland Maria Kotitsa, Department of Psychology, Institute of Psychiatry, De Crespigny Park, London SE5 8AF, UK Mairi S. MacLeod, Department of Psychological Medicine, Gartnavel Royal Hospital, 1055 Great Western Road, Glasgow G12 OHX, UK Robin Morris, Head of Neuropsychology, Neuropsychology Unit, PO Box 078 Institute of Psychiatry, De Crespigny Park, London SE5 8AF, UK

ix

x

LIST OF CONTRIBUTORS

Thomas C. Ormerod, Psychology Department, Lancaster University, Lancaster LA1 4YF, UK Adrian Owen, MRC Cognition and Brain Sciences Unit, 15 Chaucer Road, Cambridge CB2 2EF, UK Louise H. Phillips, Department of Psychology, William Guild Building, Kings College, University of Aberdeen, Aberdeen AB24 2UB, UK Mary Jo Rattermann, Department of Psychology, Franklin and Marshall College, PO Box 3003, Lancaster, Pennsylvania 17064-3003, USA Jon S. Simons, Institute of Cognitive Neuroscience and Department of Psychology, University College London, 12 Queen Square, London WC1N 3AR, UK Lee Spector, Dean of the School of Cognitive Science, Hampshire College, Amherst, MA 01002, USA Geoff Ward, Reader, Department of Psychology, University of Essex, Wivenhoe Park, Colchester, Essex CO4 3SQ, UK

CHAPTER ONE

Introduction to the psychology of planning Geoff Ward Department of Psychology, University of Essex, UK

Robin Morris Neuropsychology Unit, Institute of Psychiatry, London

INTRODUCTION The terms “plans” and “planning” can be used to refer to many different aspects of cognition and cognitive control in everyday life. One use of the term “plan” is to describe a procedure for achieving a particular goal or desired outcome. For example, when we ask a friend or colleague “So, what’s the plan?” we are often hoping for a set of directions to guide our thoughts and actions. That is, we are hoping for directions on what to do and when to do it, and this in turn might tell us those things that are most important and those things to watch out for. Ideally, the plan that is shared should be complete (that is, the contents and their ordering satisfactorily accomplishes the goal), efficient (the component thoughts and actions should hopefully have been evaluated and optimized), and foolproof (the instructions should be easy to memorize, monitor and execute, with little chance of things going wrong). However, in everyday life, plans may still be useful without offering explicit guidance or instructions. A map of the London Underground or an architect’s diagram detailing the layout of a house may also be properly referred to as plans, but they provide a representation or overview of a project or problem, rather than a set of directions. This type of plan refers to the appropriate organization of knowledge, and facilitates integration of the component parts of a problem, allows for mental simulation to generate and evaluate new ideas, and affords increased understanding to test and detect problems before they occur. That is, although these plans may themselves be 1

2

WARD AND MORRIS

the result of prior planning, they additionally offer a representation or framework of a problem in such a way that helps promote future plans and planning. For example, a clear understanding of the current state of play and what needs to be done, together with a knowledge of the different means and methods at our disposal may greatly facilitate the efficiency of achieving our goals. Through using planning and problem-solving tips and strategies, a selected subset of (hopefully useful) new alternatives may be discovered more quickly and more efficiently than would be the case if one used only pure trial or error, or if one systematically explored each and every possibility in turn. In psychological research on planning, the different aspects of plans and planning suggested by the preceding paragraph are given rather different names (for a fuller discussion on the different definitions and components of planning in psychological research, see Scholnick & Friedman, 1987). Many theories of planning propose that the first stage is to form a suitable mental representation of the problem (sometimes known as an image, problem space or blackboard). The representation may include the initial state and the goal state as well as a range of possible actions that could be taken. Planning tips, strategies and tactics are known as heuristics and algorithms, which can be used to work forwards from the current state of play and backwards from the goal to help evaluate and select alternative component actions. A sequence of activities or a plan of action may then be stored in working memory, a shortterm memory for temporarily holding and manipulating information, and the continued success of the plan may be monitored and evaluated by executive control mechanisms that may troubleshoot when things go wrong. Plans intended for ongoing or less immediate actions are often learned as part of our long-term repertoire of skills and facts, but, to the extent that planning involves intended actions to be taken sometime in the future, motivational and attentional control may be needed in order to carry them out appropriately. This introduction to the psychology of planning is subdivided into three sections: first, a number of key theoretical accounts of planning are introduced; second, the most important methodologies used in planning research are introduced; third, short summary descriptions are provided outlining the main themes and issues to emerge from the subsequent chapters contributed to this volume.

THEORETICAL BACKGROUND Overview A number of different theories of plans and planning are outlined in the following subsections. It is worth noting, at the outset, that these different accounts have been developed with rather different aims in mind. The first

1. INTRODUCTION TO PSYCHOLOGY OF PLANNING

3

subsection describes the influential framework for planning and problem solving developed by Simon (1978) and colleagues (Newell & Simon, 1972; Newell, Shaw & Simon, 1958). Their accounts emphasize the idea that planning can be likened to a search through a space of connected problem states, with the efficiency of the search improved by using a range of different heuristics to think forwards from the given information of a problem and backwards from the goal of a problem. The second subsection discusses two complementary approaches to planning that have emerged, partly in reaction to Newell and Simon’s (1972) account: namely whether one should create a full working plan in advance, or plan opportunistically, “on-line” as the current situation dictates. The section considers a number of different “top-down” and “bottom-up” planning and problem-solving strategies, and discusses the contributions to planning of Sacerdoti (1974, 1977) and Hayes-Roth and Hayes-Roth (1979). In the third subsection, the pioneering, human information-processing account of Miller, Galanter, and Pribram (1960) is discussed in some detail. This account was heavily influenced by the early work of Newell et al. (1958), and viewed plans as providing the essential connection between knowledge, evaluation, and action. The account was applied to a wide range of different specialist areas of psychology, including willed behaviour, language comprehension, language production, memory, and problem solving, such that their account of “Plans and the structure of behavior” provides one of the earliest general accounts of human cognition. One impact of Miller et al.’s (1960) account is that it provided one of the first general theories of human behaviour. The fourth subsection outlines a number of more contemporary general accounts of human cognition that have similarly contributed to research and discussion on planning (e.g., ACT-R, Anderson, 1993; and SOAR, Newell, 1990). A second impact of Miller et al.’s (1960) account is that it introduced a number of themes and terms that have influenced many subsequent accounts of planning. For example, they argued that knowledge and behaviour was hierarchically organized, and that “plans” were the basic building blocks of action. Miller et al. also introduced the term “working memory” to cognitive psychology in relation to holding and manipulating components of current plans and planning. They discussed the willed control of behaviour in executing plans and resolving conflicts between competing actions. These themes are addressed in a number of subsequent subsections that discuss, respectively, plans and schematized knowledge, the role of working memory in planning, and the executive control of thought and action. The final subsection recognizes the potential importance of motivation, emotional, and cross-cultural factors that have been largely ignored by theories of planning in cognitive psychology. It also highlights the fruitful discussions of planning that have recently emerged from within a developmental

4

WARD AND MORRIS

perspective (Friedman & Scholnick, 1998a; Friedman, Scholnick, & Cocking, 1987).

Planning and human problem solving This section examines the research programme of Simon and colleagues (1978; see also Newell & Simon, 1972; Newell et al., 1958) who have contributed one of the most influential human information-processing accounts of planning within a framework for human problem solving. Simon (1978) characterizes problem solving as an interaction between a participant and a task environment (the name given to the problem as presented by the experimenter). This interaction results in the participant generating a mental representation of the problem which is called the participant’s internal problem space. Participants are assumed to be highly malleable and adaptive such that only a few basic characteristics of the human information-processing architecture affect problem solving. The most important characteristics for our purposes are: (1) that problems are represented symbolically; (2) that problem solving occurs serially or sequentially; (3) that there is a limited capacity short-term memory system for holding and manipulating information; (4) that there is a virtually unlimited long-term memory system. By contrast, the participant’s internal problem space is predominantly constrained by the underlying basic structure of the problem. In a welldefined transformation problem, such as the three-disc Tower of London (TOL) task which is popular with planning researchers, the participant is presented with all the information that is needed to solve the puzzle, including the initial state, the goal state, and the operators (the range of possible options available to the participant) and their restrictions. A representative TOL puzzle (taken from Shallice, 1982, 1988) is illustrated in Figure 1.1. The puzzle consists of three coloured balls (red, green, and blue) arranged on three pegs of different heights. The pegs can hold (from left to right) a maximum of three balls, two balls, or one ball at any moment in time, and participants must convert an initial or start configuration of balls into a final or goal configuration of balls in the fewest moves. In puzzles of this sort, there are only a finite number of legal states of play, and each state can be transformed into only a small subset of other states in a single move. The overall basic structure of a problem can be achieved by constructing a state space diagram (an example is provided in Figure 1.2), which shows all the logically permissible states and how they are interconnected. A state–space diagram is useful in understanding the difficulty of a particular problem since it makes explicit the minimum number of moves to solution, the range of alternative suboptimal routes that are available, the number of deadends and loops back to the start. Planning and


5

Figure 1.1 The three-disc Tower of London (TOL) task, of the type used by Shallice (1982). The three letters indicate different coloured balls threaded onto the rods of differing height. The aim is to rearrange the position of the balls from the initial position to the goal position. Only one ball can be moved at a time.

problem-solving activity can then be formalized as the solution paths that are considered or taken by participants through the interconnected states in the problem-space. An inspection of Figure 1.2 reveals that there are 36 different states in the three-disc TOL task, with each state connected to between two and four other states. Interestingly, the overall structure of the state–space diagram can be described as three clusters of ten states, with each cluster connected to each other by two further states. States within a cluster share a common coloured ball on the lowest position on the highest peg, whereas the six states connecting the different clusters consist of the six arrangements where there are no balls on the highest peg. Despite the complex nature of the basic problem space, participants normally traverse through only a small subset of possible state spaces whilst solving a particular problem. This is because the participants do not behave randomly, nor do they systematically explore of all the possible legal states. Rather, planning and problem solving can be characterized as a search through the problem space, and a range of different heuristics (or “rules of thumb”) help limit the size of the search. For example, the heuristic of hill climbing can be used when planning forwards from a given problem state to determine which move will best reduce the difference between the current state and the goal state (that is, to test whether moves will help the participant in “getting warmer” or closer to the solution). Progress could be measured in terms of the number of balls in their correct position, or more speculatively, the removal of obstructions to balls or their goal locations. Alternatively, participants might use the heuristic of means–ends analysis (discussed in more detail, later) to work backwards from the goal state, in an attempt to break the overall goal into a sequence of more tractable subgoals. Furthermore, participants might “back up” their search, if a solution path appears unfruitful and try again from an alternative state.

Figure 1.2 The state–space diagram for the three-disc TOL task. The state space is all the possible arrangements of the three coloured balls on the rods. The arrows show how the arrangements can be transformed one move at a time to travel around the state space.


7

One of the most influential aspects of Simon and colleagues’ work is that their accounts of problem solving can be formalized as computer simulations using production system architectures (e.g., Newell & Simon, 1972). Production systems provide an architecture that offers some parallels with cognitive counterparts in the human information-processing system. They consist of a short-term or working memory, an activation memory (analogous to pattern recognition), and a long-term memory of production rules (which can be likened to condition–action statements, such as IF THEN ). Actions may change external states (as in the case of selecting a particular move) or internal, mental states (as in the addition or deletion of elements in working memory). When given conditions activate incompatible actions, or when two or more different conditions compete, then a set of conflict resolution rules helps select the most appropriate action, based on the recency, the specificity, or the success of past executions of that rule. Production system models have been influential because they explicitly describe the strategies (the contents of the production rules), and what is thought (the content of working memory) at different stages in planning and problem solving. The models can then be compared with the actual actions that participants make, as well as their verbal descriptions or verbal protocols (Newell & Simon, 1972) whilst problem solving. This theorizing has been applied to the Tower of Hanoi (TOH) test, a procedure similar to the TOL, but requiring the movement of discs of differing sizes, with the constraint that a larger disc cannot be placed on a small one (see Figure 1.3). Furthermore, in the case of the TOH problem, production systems have allowed a number of different hypothetical strategies to the same puzzle to be formalized (Simon, 1975). The TOH consists of three equal height pegs and a number of discs, typically between three and five of different sizes. The aim of this puzzle is to move the discs from the initial state (in which a tower of discs is on the left-hand peg, with the smallest on the top) to the goal state (in which the discs are similarly arranged on the right-hand peg) in the minimum number of moves, by moving one disc at a time, and observing the rule that larger discs may not be placed on top of smaller discs. In addition, production systems have also been used to model explicitly the successive strategies that participants acquire through learning over repeated presentations of the same problem (Anzai & Simon, 1979). The problem-solving account of Simon and his colleagues has also formalized what the participant comprehends or understands from the task as presented by the experimenter. Changes in the cover story can greatly affect the ease of learning and applying the rules (e.g., Kotovsky, Hayes, & Simon, 1985), even for puzzles of the same underlying basic structure. It is worth spending some time at this point detailing some of the history of Simon and colleagues’ account, to illustrate the types of problems that

8

WARD AND MORRIS

Figure 1.3 A three-disc version of the TOH. The aim is to move the discs of differing sizes from the initial position to the goal position. Only one disc at a time can be moved and a larger disc cannot be placed on a smaller one.

have been addressed, and examine the types of programs that were created. In one of their earliest papers, Newell et al. (1958) described an informationprocessing program called Logical Theorist that was capable of discovering proofs for theorems in logic. Logic Theorist solved these problems by searching for possible solutions, generating possibilities out of the given facts and the outcomes of completed and partial solutions to prior problems, and evaluating the alternatives before proceeding. The program was implemented using a computer, thus eliminating any sense of “mystery” to the problemsolving process. The program incorporated the basic rules of symbolic logic and a list of expressions (axioms taken from a classic’s textbook and previously learned theorems). It attempted to solve 52 theorems taken from an early chapter of the same book. These were attempted in the same order as in the text, and the program completed 38 of these problems successfully, with the time spent trying to solve each problem related to the length of the proof. Newell et al. (1958) showed that Logic Theorist made use of the solutions to prior problems, by demonstrating that the twelfth problem, which had been relatively easy to solve having completed prior solutions, could not be solved if it was encountered first (when Logic Theorist could make use of no prior solutions), and took longer to solve when encountered second (when it had previously solved only a single selected theorem).


9

An examination of the way Logic Theorist had attempted the problems suggested that its performance could be characterized as showing mental set (e.g., the directional nature of applying different proof methods), insight (in as much as there was a suddenness of discovery accompanied by a sudden grasp of the structure of the problem), and hierarchical behaviour (in that problems were deconstructed into subproblems). In fact, four proof methods – substitution, detachment, forward chaining and backward chaining – were applied in that order to the initial “given” information and were subsequently also applied to the partial solutions obtained from applying these methods. The proof methods shared common subprocesses such as matching and similarity testing, and there was an overarching executive process implemented to coordinate the use of these methods. Trial and error use of the methods was reduced by using a number of heuristics: applying the methods in an order determined by trying manipulations involving variables that appear in the problem expression; working backwards from the problem expression; and searching the list of learned theorems for “similar” theorems. When trial and error attempts occurred, they did so in a problem space of possible solutions, and a strategy was used that permitted the search to be limited to a small subspace of alternatives, or generated elements in the space in an order that made early discovery more probable. Logic Theorist was just one of a number of different informationprocessing systems that were discussed by Newell and Simon (1972) for solving specific problems, and others included those for solving cryptarithmetic puzzles and playing chess. However, Newell and Simon also discussed the General Problem Solver, a problem-solving program that was designed to be independent of a specific task. Like Logic Theorist, the General Problem Solver attempted to solve problems by reducing the space of alternative solutions through which the program searched. However, whereas Logic Theorist used methods such as working backwards from the goal, the General Problem Solver used a number of more powerful heuristics, including means–ends analysis and the planning method. Means–ends analysis is a heuristic that first detects differences between the available object and the desired object, and then tries methods for transforming features of the object that help to reduce that difference. The heuristic can therefore be likened to hill climbing (described earlier), which also tries to reduce the difference between the current and goal states. However, a key difference is that with hill climbing, if a desired operator cannot be applied to reduce the difference, no action is taken. By contrast, with means–ends analysis, if a desired operator cannot be applied to reduce the initial difference then a new subgoal is set up to reduce that goal, and the process of means–ends analysis is applied recursively until either an operator can be performed to achieve a lowest level subgoal or failure occurs. Furthermore, it may be profitable to modify the objects so that the desired operator may now be applied, and it may be profitable to reduce

10

WARD AND MORRIS

some “difficult” differences even if this means introducing additional (but “easier”) differences. The planning method is a heuristic for creating simpler problems by omitting certain details of the original objects and operators, and then generating and solving the simpler more abstract problem. This solution could then be used as a plan to solve the original problem. The Newell and Simon (1972) account has been developed further, as detailed above, by Simon and colleagues (1978; Kotovsky et al., 1985), and has been highly influential in the development of more contemporary accounts of cognition, discussed in a later section, such as SOAR (Newell, 1990) and ACT-R (Anderson, 1993). The influence of this line of work can also be seen in many of the contributions in this volume (e.g., Davies, chapter 2; Gilhooly, chapter 4; Ward, chapter 5; Phillips, Macleod, & Kliegel, chapter 6, amongst others) who have continued to use well-defined problems as an appropriate method for studying planning processes.

“Top-down” and “bottom-up” strategies in planning research The success of Newell and Simon’s (1972) General Problem Solver has led to a wide range of different planning and problem-solving strategies in human and artificial intelligence (AI) systems. Hoc (1988) has characterized these planning strategies as “top down” and “bottom up” (for a discussion of alternative nomenclatures, including the new distinction between “initial” and concurrent planning, see Davies, chapter 2, this volume; for the additional distinction between “local” and “global” planning, see Ormerod, chapter 3, this volume). Top-down strategies refer to processes in which “higher order” knowledge guides, moderates, and (hopefully) enhances more basic decision making. One difficulty with the most elementary top-down strategies (such as those relying entirely on means–ends analysis) is that the planner does not have a good source of information as to which operators will best solve the problem. If only a blind search of the problem space is performed, then as the length of the plan increases, so the number of alternative multistage plans will increase combinatorially. To limit the search, additional heuristics are needed. As we have seen in the preceding section, Newell and Simon’s (1972) General Problem Solver used the planning method to constrain the problem search: a more abstract plan using simple operators is first generated and then this solution is successively refined such that an executable procedure is found. Similarly, Sacerdoti’s (1974) ABSTRIPS problem solver constrains the search space by solving the problem using simpler, less constrained operators. This creates a hierarchical planning structure in which there are abstract, high-level plans consisting of a series of states, stepping stones, or “defined islands” through which a solution must proceed. The lower level detailed plans can then be


11

implemented by finding solutions to the smaller problems of navigating between the islands. Although ABSTRIPS can considerably reduce the problem of combinatorial explosion, the formation of a detailed plan is likely to be unsuccessful if the higher level plan is incorrect. One example of a poorly constructed higher level plan is a solution to the goal to “paint the ceiling and paint the ladder” described by Sacerdoti (1977) that involves forming a plan with two parallel subgoals of “painting the ladder” and “painting the ceiling”. Clearly, execution of this poorly ordered plan will involve getting paint from the ladder on one’s hands as the ceiling is painted. One alternative approach used by Sussman’s (1975) HACKER is to produce a “first guess” by solving a conjunction of goals in a linear fashion, and then recover from any errors to produce a correct solution. A later hierarchical planning model by Sacerdoti (1977) called NOAH successfully plans to “paint the ceiling paint the ladder”. It uses a “least commitment” strategy that tries not to overconstrain the order of the operators too early in the planning process. This is achieved by first expanding the most detailed plan currently available and then evaluating, refining and optimizing the plan using constructive critics to perform actions such as resolve conflicts, reorder components, and eliminate redundant operations. In the case of the goal to “paint the ceiling paint the ladder”, it would involve planning first with “painting the ceiling” before starting with “painting the ladder”. By contrast, Hoc (1988) characterizes other planning strategies as “bottom-up” strategies. These strategies refer to planning processes which are driven by incoming data. Two examples are plan recovery, in which a prelearned action sequence is elicited by a triggering stimulus or by reasoning by analogy, and plan revision, in which detailed evidence gathered from on-line performance shows an original plan to be inadequate, with the result that the plan is modified to take into account the new data. Although many AI models of planning such as Sacerdoti (1974) have emphasized top-down planning, Hayes-Roth and Hayes-Roth (1979) highlighted the fact that humans make many planning decisions “on-line”, such that many decisions may be based largely on our current thoughts and observations. Planning can therefore be characterized as both bottom up and top down, incremental or partial instead of complete, and hetararchical rather than hierarchical in nature. They illustrated this assumption by asking five different participants to generate verbal protocols whilst planning six different errand-planning tasks. For example, the individuals were presented with a fictitious street map of a central shopping district and a list of things to buy and do, and participants thought out loud whilst saying how they would organize their time. They observed that the participants did not plan in a strictly forwards manner, from major goal through to minor subgoal, but rather planned different parts of the sequence at different points. Decisions at a given point affected those earlier and those later in the plan, and also

12

WARD AND MORRIS

affected those at a more abstract and more detailed level. At times, the planning simulated real-life performance, with the results of the simulation guiding how future plans develop, both in terms of what can now be achieved (event-driven process) and when these actions will be achieved (time-based processes). Hayes-Roth and Hayes-Roth used their protocol analyses to formalize a cognitive model of planning. At the centre of the model is a common blackboard, where the products of tentative decisions are recorded. The blackboard is divided into five planes, and these planes are further divided into different levels of abstraction: the Plan Plane represents those actions that are currently intended; the Plan-Abstraction Plane represents desired aspects of plans; the Knowledge-Base Plane represents observed and calculated facts; the Executive Plane represents the priority for allocating resources; the MetaPlan Plane represents the planner’s understanding of the problem, the available operators, and evaluation criteria. Planning decisions are made by many independent cognitive specialists, who can read relevant information from any area of the blackboard. If their conditions are satisfied, then they act by making tentative decisions based on this information and their own decisionmaking heuristics. The cyclic planning process, prioritized by the executive, proceeds by first selecting specialists based on whether their conditions are met, followed by the specialist executing their actions. These are recorded on the blackboard and the information may be used on subsequent cycles. When there is a complete plan that satisfies the evaluation criterion, the planning process ends, but planning may also fail under certain circumstances. A simulation of this cognitive model satisfactorily recreated certain aspects of the planning process, demonstrating the potential benefits of this opportunistic emphasis to planning. Although it is clear from this section that it is at least logically possible for our planning behaviour to be influenced by top-down and bottom-up processes, quite which planning strategy is adopted by a participant at any moment in time may be affected by working memory limitations (see Gilhooly, chapter 4, this volume) and also by the complexity of the problem, the task environment of the problem, and additional individual and group differences (see Davies, chapter 2, this volume, for a full discussion).

Plans and the structure of behaviour The section reviews one of the earliest and most influential cognitive accounts of plans and planning, the account provided by Miller et al. (1960) in their seminal monograph, Plans and the Structure of Behavior. Miller et al. defined a plan as “any hierarchical process in the organism that can control the order in which a sequence of operations is to be performed” (p. 16), and argued that all plans consist of one or more Test Operate Test


13

Exit (TOTE) feedback loops that are responsible for coordinated and organized acts. Figure 1.4 illustrates a simple plan that consists of a single TOTE feedback loop. In this example, the plan is to turn on the television (TV), and this will be achieved by pressing the “on” button on the TV remote (controller). The loop consists of Test TV (Test that the TV is on) that must be compared with the current state of affairs. Let’s say that the TV isn’t on (and so the conditions are not yet satisfied). If the Test conditions were satisfied then the loop is Exited, but if, as in our case, the Test conditions are not yet met, then an Operator (some consequential action or mental activity) is performed. In our case, the Operator is (Press green button on the TV remote). Once the operator has been applied (that is, the green button on the remote pressed) the current status is re-tested. If the TV is still not on (as sometimes seems to happen with some less reliable TV remotes) then the operator is reapplied until such time as the conditions are finally satisfied, at which point the TV will be on and the plan can be exited. Although this terminology may seem somewhat arcane, the function of the TOTE is relatively simple. A plan is like an intention to achieve a desired goal by performing an action, with the additional feature that, through feedback, the action is repeated until the intended goal has been achieved. The plan described in Figure 1.4 is very simple, consisting of only a single TOTE unit. However, more complex plans can be devised that consist of more than one TOTE that are arranged in a hierarchical manner. Miller et al. provide the example of hammering a nail such that it is flush. Figure 1.5 illustrates this hierarchy. The top TOTE consists of the Test to see whether the nail is flush. If the nail is indeed flush, then the TOTE is exited and the plan is accomplished. If it is not, then the operator HAMMER is performed, and the procedure repeated until the nail is flush. Now the operator

Figure 1.4

The TOTE feedback loop for turning on the television using a remote controller.

14

WARD AND MORRIS

Figure 1.5

A more complex hierarchy of TOTEs showing the plans for hammering a nail.

HAMMER can be conceived of as two separate TOTEs: a TOTE that applies when the hammer is found to be up (the operator “striking the nail” is performed) and a TOTE that applies when the hammer is found to be down (the operator “lifting the hammer” is performed). This more complex plan reveals how a series of TOTEs can be organized into ever-more complex plans using the same building blocks of plans at each level. A number of features of Miller et al.’s account are worth highlighting as they have been influential in future developments within plans and planning. First, as illustrated above, Miller et al. proposed that our knowledge is highly structured, and our plans are often hierarchically organized, with superordinate level plans activating multiple component subordinate plans. Plans could therefore be properly seen as the interconnection between our knowledge and our behaviour. Second, Miller et al. proposed a quick-access “working memory” to provide a special state in consciousness in which currently executed plans were taken out of “dead storage” and placed in control of information capacity. Working memory was argued to store necessary data and components of plans, such as the current values and the remaining intentions of the active plans, which could be used to determine how far through the plan the person has proceeded, and could be used to resume an interrupted plan. Miller et al. also asserted that an “effort of will” may be necessary to properly control our behaviour. Quite which plan we adopt at any given time was assumed to be governed by our motives, which were defined as consisting of values and intentions. Values were hypothesized to be a class of test conditions that may be specified in TOTE units, whereas intentions were thought


15

to be those behaviours that were yet to be performed from an initiated plan. Due to the complex nature of some grand plans, in which there are hierarchical structures of TOTE units, there will be a hierarchical structure of values. To satisfactorily perform the grand plan, some component subgoals might have negative evaluations associated with them, such that a person may properly intend and then carry out actions that they consider undesirable. Thus, “higher order” TOTE units may need to override evaluations from “lower level” TOTE units, and if unexpected or not fully thought through TOTE units produce too much negative evaluation, then the grand plan may be suspended, due to a change in the value of the highest TOTE units. They argued that emphatic inner speech could be an integral part of the plan controlling structured behaviours. Failures in the control of behaviour were assumed to occur because plans must compete with other plans to advance several plans simultaneously. Forgetting intentions may be due to competition with other plans, working memory failure, and due to the value of a plan being reduced upon re-evaluation, or through changes in external conditions. Thus, external memory aids for recording plans, their intentions and progress should facilitate plans. Miller et al. withheld a discussion of formulating new plans until the penultimate chapter of their book. They assumed that many new plans are learned by imitation or instruction. However, they also proposed that many plans for solving “higher order”, novel and complex problem solving may be performed by taught meta-plans: plans for generating plans may be domainspecific heuristics, such as those used for strategies when playing chess, or may be domain-general heuristics, such as means–ends analysis. They anticipated some criticism for this rather brief discussion of novel and complex planning, but they were heavily influenced by the pioneering breakthroughs in cognitive science performed by Newell et al. (e.g., Newell et al., 1958), and so they could direct their critics to fully working simulations where this approach had been proven to be successful: computer programs using an architecture of production rules which could solve geometry puzzles and play chess. They were particularly encouraged because the TOTE units that they discussed throughout their treatise resembled the production rules used by Newell et al. for testing whether certain conditions were met and then performing actions accordingly.

Plans and schematized knowledge One influential aspect of Miller et al.’s account of plans and planning is that plans are best considered within a framework of understanding the structure of behaviour. Central to this understanding was that behaviour and knowledge were hierarchically organized, and that more complex plans could be composed from more complex arrangements of simple building blocks. In

16

WARD AND MORRIS

this section, we outline Schank and Abelson’s (1977) account of scripts, plans, goals and understanding, which also focuses on understanding the organization and structure of behaviour. Schank and Abelson were interested in the way humans and computers could learn and organize knowledge, and in the ways this organized knowledge was used to aid understanding. They argued that we use a mixture of two classes of knowledge to help us understand (and in some senses predict) another person’s actions. Specific knowledge refers to detailed knowledge for frequently encountered events and routines, such as going to the doctors, going to a theatre, or going to a restaurant. Typically, specific knowledge consists of standard sequences or causal chains of component events, termed scripts, which could be used to infer missing details when a participant tries to comprehend an incomplete example of a standard event. For example, a typical script for entering a restaurant might contain the components “entering”, “ordering”, “eating” and “leaving”, and each component may consist of many specific actions. Thus, “entering” may consist of “walking into the restaurant”, “look for a table”, “decide where to sit”, “go to table”, “sit down”, etc. We can therefore infer these details (unless we hear otherwise) as default actions that are likely to have occurred in a discussed event. Furthermore, our knowledge of restaurants might be such that we can understand subtle variants of the restaurant script depending upon whether we are told that the restaurant is a family steakhouse restaurant, a fast-food restaurant, a chic French bistro, or serves Indian cuisine. Even within these subtle variants, an understanding of one piece of information (e.g., the prices of the dishes on the menu) might allow one to infer other pieces of more or less important information (e.g., the requirement to make a reservation, the likely dress code, the appropriateness of tipping, the quality of the napkins, etc.). According to Schank and Abelson, general knowledge in the form of named plans is required in order to understand how humans perform general actions that satisfy their needs. Such plans include those satisfying biological needs, those relating to enjoyment or relaxation, those for achieving a valued possession or social position, or preserving or improving the health or wealth of people. Each of these plans may have pre-conditions that suggest attaining instrumental goals when the method for achieving these pre-conditions is known or suggest inspecting a set of alternative actions that could be used to satisfy a goal. Furthermore, Schank and Abelson proposed crisis goals to deal with serious or imminent threats, and these were assumed to take precedence. Furthermore, Schank and Abelson assume that plan understanding is needed to help understand novel, unexpected, or less frequently experienced events and situations. In each of these situations, by finding a plan (or by creating a plan, in the case of novel plans), a person can infer the reason for another person’s behaviour and can use this to help make sense of what they


17

are doing. Plans may be complicated and involve a main goal that is achieved by using several simpler goals (and their associated methods of attainment) as building blocks. An individual action may then be properly understood as an element within the structure of the main plan. It is interesting to note that there are similarities between the aims of Schank and Abelson and Miller, Galanter, and Pribram. Both sets of workers were interested in understanding the structure of behaviour, both considered plans to be ordered lists or sequences of component actions that satisfy goals, and both saw plans as necessary building blocks, organized in hierarchical structures. In addition, it could also be argued that both sets of researchers have contributed more to the use and long-term structure of existing plans, rather than the creation of new plans through problem solving (cf. Newell et al., 1958; Newell & Simon, 1972). However, whereas Miller, Galanter, and Pribram have emphasized the executive control of plan execution, Schank and Abelson have emphasized the higher knowledge structures of scripts in plan comprehension and have attempted to formalize this structure through computer simulation. The overall claim that our planning and understanding is affected by our schematized knowledge is still to be found in more contemporary accounts of differences between novices and experts. For example, Davies (chapter 2, this volume) discusses the role of expert knowledge with relation to the adoption of different planning strategies; Ormerod (chapter 3, this volume) considers expertise with relation to differences in planning and creative design; and Gilhooly (chapter 4, this volume) discuss differences in planning by expert and novice chess players.

Planning and more general theories of cognition The account of Miller et al. (1960) provided a description of plans and planning within a more or less complete general theory of cognitive psychology, that considered the relationship between schematized knowledge, working memory, planning heuristics and executive control, and applied the account to a wide range of cognitive activities. Miller et al.’s account identifies one difficulty in constructing and evaluating theories of planning. That is, one’s theory of planning is inevitably dependent upon one’s view of contributing cognitive mechanisms, such as working memory, executive control, and the representation of knowledge. Thus, to construct and evaluate a complete theory of plans and planning is, to some extent, to construct and evaluate a complete theory of cognition. Although no truly complete theory of cognition currently exists, a number of authors have specified the mechanisms for the mental representation of knowledge, working memory and executive control within different broad theoretical frameworks of cognition, and then have used these theories to discuss or investigate plans and planning.

18

WARD AND MORRIS

The SOAR architecture (Newell, 1990) provides a candidate unified theory of cognition. Derived from the production-system architectures of Newell and Simon (1972), SOAR considers all cognitive activity as transformations of states by operators within a state space. The currently active information pertaining to the current state is stored in SOAR’s dynamic memory, akin to a working memory. Permanent knowledge is stored as production rules (discussed earlier, with reference to Newell & Simon, 1972), and if the conditions of the rules are satisfied by the appropriate terms in SOAR’s dynamic memory, then all the respective actions are taken in parallel. These actions may include changes to the contents of SOAR’s dynamic memory. One difficulty that SOAR must overcome is the problem of dealing with scheduling conflicting or competing actions. Under these circumstances, where there is uncertainty as to which action to undertake an impasse is reached, which is resolved through the creation of a subgoal specifically for that purpose. Newell (1990) used a planning and problem-solving environment known as the blocks world to illustrate the resolution of an impasse through the creation of a hierarchy of subgoal (Figure 1.6). In the blocks world, there are three blocks sitting on top of a table, and each block can be put on top of every other block. An example blocks world problem is illustrated in Figure 1.6(a). The blocks world is represented in its own problem space in SOAR, consisting of the initial state, the desired state, and the state to which the current state would be transformed if different alternative operators (or actions) were selected and performed. Planning and problem solving consists of searching the problem space for the move that will best achieve the goal. This occurs through a decision cycle in which information about activated production rules is first accumulated and a preferred decision as to which to execute is then applied, based on the preference values associated with the productions. If one choice is unequivocally preferred then this is taken, if no operator is available or if more than one operator appears equally preferable then a state of impasse is reached. A state is then set up (which may be referred to as a substate when compared to the original superstate) which has the subgoal to further search SOAR’s knowledge base in an attempt to find information that will resolve the impasse. Such information may consist of a suitable operator when one was previously not available; or in the event of equally preferable alternatives, the newly acquired information may make one operator seem more preferable than another. Such a search may itself run into difficulties and result in a further impasse, which itself can be resolved by establishing a further state (in effect a “sub-substate”) to find additional information to break the deadlock. Resolving impasses may therefore require the recursive creation of a whole hierarchy or stack of spaces. If the impasse is resolved, then a new production rule is learned that combines the conditions in the superstate together with those items in dynamic memory that are causally implicated in resolv-


19

Figure 1.6 (a) A puzzle from the blocks world environment explored by SOAR (Newell, 1990); (b) a puzzle from the five-disc TOH task used by ACT-R (Anderson, 1993).

ing the impasse to the resolved actions. This new rule will resolve that impasse situation should the same combination of events occur in the future. The ACT-R architecture (Anderson, 1993) provides a second candidate model of cognition. It represents long-term knowledge in two symbolic systems. Like SOAR, procedural knowledge is represented as production rules, but, additionally, declarative knowledge is represented as a network of interconnected nodes. Anderson (1993) has discussed planning and problemsolving behaviour on a range of different tasks to illustrate different features of the ACT-R architecture. For example, the five-disc TOH puzzle was used to exemplify the push-down, pop-up goal stack structure of ACT-R, which is used to remember intentions when problem solving. In the standard five-disc TOH, the start state would have all five discs in descending order of size (smallest on the top) on the left-hand peg, and the goal would be to move all the discs to a similar tower on the right-hand peg. The puzzle can be broken down into a series of subgoals using the goal recursive strategy (described briefly, above, in the account of Simon, 1975). Initially, a subgoal can be created which sets to move the largest disc from the left-hand peg to the righthand peg. This can only be achieved by moving the four smaller discs into a four-disc tower on the middle peg. Creating a tower of four discs on the

20

WARD AND MORRIS

middle peg can itself be achieved by creating a three-disc tower on the righthand peg, which in turn can be achieved by creating a two-disc tower on the middle peg, and so on. Thus, the TOH can be conceptualized as the setting (or pushing) of a series of subgoals or goal pushes to create ever-decreasing towers on alternate pegs to allow the component discs in the tower to be moved. ACT-R predicts that the processing time should increase with the setting of goal pushes. This prediction was tested by comparing data obtained from human participants who were trained in the goal recursive strategy with ACT-R simulations. Participants were given a range of five-disc TOH puzzles (such as an example shown in Figure 1.6(b)), which differed in their arrangement of discs in the start states and the goal states. In line with the predictions of ACT-R, the “planning time” increased with the number of predicted goal pushes, highlighting that the setting of subgoals can take processing time to establish. Planning and problem solving using ACT-R was also explored in a navigation task. In this task, the participants’ aim was to manoeuvre from a start location to a goal location using a schematic computer map that offered only partial knowledge pertaining to the destination of the different alternative routes. As predicted by ACT-R, the time participants took to make decisions at road junctions increased almost linearly with the number of productions considered by the model. Anderson (1993) also applies ACT-R to other cognitive skills such as computer programming and geometry problems.

Planning and working memory A second influential aspect of Miller et al.’s account of plans and planning was the proposal that a quick-access working memory was necessary in order to temporarily maintain the products of currently retrieved plans. The term working memory is now a widely established term used by a range of different theorists to describe many different specific theories (see e.g., Miyake and Shah, 1999). Central to these accounts is an attempt to understand the striking memory limitations that often arise when we are asked to maintain and manipulate even relatively small quantities of information. Perhaps the most influential account of working memory is the multi-component view of working memory proposed by Baddeley (1986; Baddeley & Hitch, 1974). In Baddeley’s (1986) working memory model, three short-term memory systems are posited to store and manipulate information that is temporarily highly available whilst performing cognitive tasks. The model consists of the phonological loop for processing and maintaining verbal material, a visuospatial sketchpad for use with visuo-spatial materials and a central executive


21

that is considered to be an attentional coordinator and controller. The model has been applied to a wide range of different cognitive tasks, such as learning, reasoning and comprehension (see e.g., Baddeley and Hitch, 1974; Andrade, 2001). One approach used by its protagonists is to examine the patterns of interference of different concurrent secondary tasks on the performance of different primary cognitive tasks of interest. The phonological loop component is assumed to be selectively impaired by concurrent articulatory suppression (preventing articulatory rehearsal by requiring participants to say “the-the-the” continuously whilst performing the primary task). The visuospatial sketchpad may be impaired by concurrent tapping, whereas the central executive is selectively impaired by attentionally demanding tasks, such as random generation. Baddeley (1990) has discussed the application of the working memory model to planning tasks, such as playing chess. He reports the results from an unpublished study by Bradley, Hudson, Robbins, and Baddeley, who investigated strong and weak chess players’ abilities to reproduce a briefly displayed chess position taken from an actual game between master chess players, with or without different secondary tasks. This research showed that strong players were more accurate than weak players at reproducing the chessboard positions, and that both groups showed similar effects of the secondary tasks: concurrent random tapping and concurrent systematic tapping significantly reduced the accuracy of the positioning of the chess pieces, whereas concurrent articulatory suppression had no effect on the accuracy compared with a control condition. Baddeley interpreted these results as suggesting that the visuo-spatial sketchpad and the central executive are used when playing chess (see also Robbins et al., 1996). Similar patterns of selective interference implicating the visuo-spatial sketchpad and the central executive in planning tasks have been found by Phillips, Wynn, Gilhooly, Della Sala, and Logie (1999), Phillips, Wynn, McPherson, and Gilhooly (2001) on the TOL task (see Gilhooly, chapter 4, this volume; and Phillips, MacLeod, & Kliegel, chapter 6, this volume, for further details). It is interesting to note that working memory in ACT-R is not limited by the capacity of a special working memory store or set of stores (Lovett, Reder, & Lebiere, 1999). Rather, the contents of working memory can be considered to be those pieces of declarative knowledge that are currently highly activated. The level of activation of a node in delarative memory is dependent upon the contextual relevance of a fact, based on the current goal of the system, as well as the time that it was last used (trace decay). Working memory limitations arise because it is assumed that there is an overall limit to the total amount of activation in the declarative system, and so this in turn sets a limit on the degree to which goal-relevant items can be differentially activated. Similarly, there is no fixed capacity limit to SOAR’s working memory or dynamic memory. Rather, the capacity of SOAR’s dynamic memory is

22

WARD AND MORRIS

unbounded, although there are working memory limitations based on similarity based interference limiting the number of similar items in dynamic memory. For example, one constraint is that SOAR can hold at most two items from the same type, e.g., two sentences (Young & Lewis, 1999). A full account of many different models of working memory, including further discussion of Baddeley’s working memory model (Baddeley & Logie, 1999), ACT-R (Lovett et al., 1999), SOAR (Young & Lewis, 1999) together with an introduction and discussion can be found in Miyake and Shah (1999).

Planning and willed (executive) control Miller et al.’s (1960) account is also influential because it proposed that executive or willed control was necessary in certain planning situations. Such situations include the difficulties in controlling planned action when there are multiple, competing plans, and when it is necessary to carry out negatively evaluated, lower level plans that are nevertheless essential in order to attain highly positively evaluated, higher order plans. The examination of planning tasks has received increased theoretical interest since the Norman and Shallice (1986) model for the control of automatic and willed action was given prominence as a candidate model of the central executive in Baddeley’s (1986) working memory model. The Norman and Shallice model is also based on the production system architecture (Newell & Simon, 1972). As with production rules, it is assumed that external and internal events may automatically trigger respective condition–action schema. Schema may be hierarchically arranged such that the activation of a top-level schema may activate lower level “sub-schema”. Multiple schema could be run off in parallel as long as the intended actions were not conflicting. They proposed two levels of control. Contention scheduling was proposed to deal with pre-learned conflicts in intended actions. In these cases, one schema is selected (by inhibiting its competitors) and the selection of the activated schemas is based on such information as the frequency and recency of that schema being executed in the past, together with the current contextual environment. In dangerous, novel or complex situations, or where troubleshooting was necessary, intervention was required by the highest level of attentional control, the supervisory attentional system (SAS). Norman and Shallice (1986) explicitly stated that novel and complex planning also required this highest level of attentional control, and planning tasks have therefore become useful exploratory tools for specifying the functioning of the executive (for a more detailed discussion of the Norman & Shallice model and its developments, see the neuropsychology subsection, below). The issue of planning and executive control is discussed more fully by Ward (chapter 5, this volume).


23

Widening the scope of cognitive planning research In this final subsection, we discuss recent advances in planning research in developmental psychology. Friedman and Scholnick (1998a) have argued persuasively that a complete account of planning should include factors that are simply ignored in many cognitive accounts. In considering planning from a developmental perspective (for related papers, see also the contributions to the edited volumes by Friedman & Scholnick, 1998b; Friedman et al., 1987), they identify additional emotional, motivational, cultural and social factors that may be critical in determining whether a participant is willing and able to plan, whether this be on their own, or with peers or family members to complete a task. Central to their approach is the need to resolve an apparent paradox: they claim that many problems associated with adolescents (e.g., teenage pregnancy, drink driving, risk-taking behaviour) are often considered to be due to a failure to plan, but these very same individuals may efficiently plan ahead when devising schemes for obtaining illicit substances or organizing social get-togethers. Such a paradox can be resolved if one acknowledges that societal, family and peer group norms all influence the cognitive, personality and motivational components of an individual’s planning ability in terms of, for example, an individual’s knowledge base, beliefs, skills and goals. Different contributors to their two edited volumes discuss the development of planning behaviour as it emerges from an understanding of anticipation, knowledge of regular sequences and temporal events in the environment, and emotional management of behaviour. Other contributors discuss interpersonal and social aspects of planning, including planning in health-related areas and in the workplace. Although some of these areas are discussed in areas of cognitive psychology (e.g., Oatley & Johnson-Laird, 1987), and social psychology (e.g., Azjen, 1985, 1991; though see also Armitage & Conner, 2001), it may be that with time many of these ideas could be usefully transferred to cognitive accounts of adult planning. More importantly, some of these contributions discuss concepts, theories, and methods that are central to the cognitive psychology of planning. For example, in Scholnick and Friedman’s (1987) discussion of the different uses of the term planning in the psychological literature, they identified six different components of goal-directed actions that are referred to by planning theorists. These include: (a) the generation of an appropriate mental representation; (b) choosing a goal or purpose; (c) deciding to plan; (d) formulating a plan; (e) executing and monitoring plans; (f) learning from the plan. They further characterized planning as an activity that occurs simultaneously on three levels. Planning occurs when dealing with a current problem or other situation, when dealing with anticipated future possible situations, and when

24

WARD AND MORRIS

controlling behaviour such that plans are implemented. Scholnick and Friedman argue that rather narrow and diverse definitions of planning exist because different researchers have tended to highlight only a subset of these different planning components and levels of activity. They further identify a number of important distinctions in different theories of planning, and these include (1) whether planning is a general cognitive skill or a context-specific activity; (2) whether planning is a mandatory activity or a voluntary strategy; (3) whether individual differences in planning are based on the number of planning components present and/or the speed of their execution. These issues are at the heart of theorizing in cognitive psychology and to some extent help determine the research methods and the research questions we pursue. Similarly, Scholnick, Friedman, and Wallner-Allen (1998) provide a comparative analysis of two types of planning tasks which will be familiar to readers of this volume: those derived from the TOH and tasks concerned with grocery shopping (such as that used by Hayes-Roth & Hayes-Roth, 1979). Scholnick et al. provide a task analysis of each type of task, and conclude that although both types of task require the construction of a sequence of actions to achieve a goal, the two tasks emphasize rather different aspects of planning. They argue that the main difficulties in planning in the TOH task are the difficulties in mentally representing the problem (due to the similarity of the different states and the rather arbitrary rules of the puzzle), and the requirement to construct a series of subgoals to obtain the larger discs (since their removal is constrained). These difficulties increase the working memory load when planning the TOH. Furthermore, goal monitoring is difficult because progression to goal is not always seen in the external representation of the puzzle. By contrast, they argue that the main difficulties in planning in the grocery shopping task are the difficulties in searching for the items (an attentional process guided by categorical and geographical knowledge), and the requirement to integrate the locations into an efficient route. Once a plan is constructed, progress is easily monitored on the grocery task. They conclude that the two tasks measure essentially different abilities, such that performance on one measure is unlikely to predict performance on the other.

METHODOLOGY AND PLANNING How planning is conceptualized is determined in part by the experimental procedures used to investigate it. The tasks define the domain of enquiry and in certain instances the theories follow to account for the phenomena created by tasks and it is important to consider in outline the various techniques alluded to in this book. Two general planning methods can be identified. The first general method can be considered to be “puzzle-based” procedures.


25

These have “boardgame-like” qualities and involve simple mechanistic procedures or equipment. They include, for example, the TOH (see Davies, chapter 2, this volume) and the variants of this procedure, the TOL (see chapter 4 by Gilhooly; 5 by Ward; 6 by Phillips, MacLeod, & Kliegel; 7 by Owen; 8 by Morris, Kotitsa, & Bramham, this volume). The procedures may be well defined with a limited problem space (as is the case for the TOH), or illdefined, as in the nine-dot and radiation problems (see chapter 3 by Omerod, this volume). Such tasks have the advantage of laboratory control. For example, the TOH has been used in numerous studies, and provided a convenient tool for functional neuroimaging. This task has the advantage that cognitive functions which contribute to performance, for example, working memory and response inhibition, can be dissected and studied in depth. One characteristic of such procedures is that they are less prone to the effects of the expertise or knowledge base of individual participants, unless, of course, they have encountered the particular puzzle before. However, it has been argued that this procedure does not elicit the type of “real-world” planning activity, and there is controversy over whether the task involves a core feature of planning, namely “look-ahead”. Nevertheless, it may be premature to dismiss the validity of these approaches, since they are tapping into mental operations or cognitive functions that may be germane to particular subsets of planning activity, for example, graphics design or using machines. Errand tasks and their variants are a second type of procedure. These involve a set of instructions or rules about completing “everyday” tasks, for example, the financial planning task by Goel, Grafman, Tajik, Gana, and Danto (1997) or the shopping tasks (described in chapter 10 by Burgess, Simons, Coates, & Channon). These have the appearance of “ecological validity” because they involve procedures familiar to most people, and they also introduce levels of complexity, analogous to everyday activity. Errand tasks may be supplemented by protocol analysis, in which the person is asked to commentate on their problem-solving activities and provide a rationale. This approach has the advantage that it may be a more rapid route to internal “cognate” processes, and guide the researcher to understanding what strategies and procedures are adopted. It also has to be considered with some caution because of the well-known phenomenon that introspection about cognition or behaviour may sometimes not be an accurate representation of what is actually happening. At first sight, this second approach might not appear promising for experimentally controlled analysis of the cognitive structures or processes that could help to explain how the mind plans. The tasks are heavily influenced by the idiosyncratic nature of personal experience. For example, on the shopping task described by Burgess et al. (chapter 10), the participant JS decided to go to the pharmacy first because “that’s the most important thing in my

26

WARD AND MORRIS

job”. Nevertheless, the approach circumvents the danger that the phenomena studied may have no relevance to everyday life; an advantage that is perhaps particularly important in neuropsychology, where the goal is practical – to understand and remediate disability. It also reflects the diversity of planning processes and how they might show dysfunction in different groups of people; a point that is illustrated by the intriguing finding that “errand” or ecologically valid tasks appear to be more sensitive to neuropsychological impairments associated with brain damage, but the reverse appears to be true for the effects of cognitive ageing. Finally, even though this approach is primarily presented as being more “practical”, the rich data it produces tends to stimulate theoretical advances in this field.

NEUROPSYCHOLOGY AND PLANNING It is possible to consider the cognitive processes involved in planning without reference to brain. Nevertheless, neuropsychological studies have not only established the neural correlates of planning, but provided theoretical advances which have informed cognitive psychology, as indicated by the section on neuropsychology in this book. The neuropsychological study of planning has a long history, and has tended to parallel the development of cognitive approaches. Specifically, the role of the frontal lobes in the organization of behaviour has been considered early on extensively in both animals (for example, Bianchi, 1895; Pavlov, 1949; Pribram, 1961) and humans (Kleist, 1934; Hebb, 1945; Luria, 1973). In particular, Luria (1969) produced insightful descriptions of disturbances in planning in patients with large lesions of the frontal lobes and has been highly influential in the important accounts of Stuss and Benson (1986), Norman and Shallice (1986, Shallice, 2002) and Grafman (chapter 9, this volume). Clinical observations of the dysregulation of behaviour have led to cognitive theorizing about the role of the frontal lobes. One approach, adopted by Stuss and Benson (1986), has been to fractionate superordinate control into specific aspects, including anticipation, goal selection, planning and monitoring, again attributing these to the functioning of the prefrontal cortex. Another approach, developed initially by Norman and Shallice (1986; see Shallice, 1988) introduced the notion of “contention scheduling”, an idea derived from artificial intelligence (AI) production systems, as discussed above, but adopted to explain clinical phenomena associated with frontal lobe damage. This model sites contention scheduling in the basal ganglia and premotor cortex (Shallice, 2002), with the SAS reliant on prefrontal cortical functioning. Here, the notion is that prefrontal cortex is more involved in nonroutine than routine operations, including the need to plan a sequence of responses when confronted with a novel situation. This theory has been developed to specify the functions of the SAS in more detail by Shallice and


27

Burgess (1996), with the Mark II Supervisory System model. Notably, this model incorporates the notion of problem orientate goal-setting processes, with monitoring and checking the results of schema implementation. In a further development of this model, the top-down modulation of schemas involves the left dorsolateral prefrontal cortex, whilst monitoring and checking involves the right dorsolateral prefrontal cortex (Shallice, 2002). The theories above in essence treat the prefrontal cortex as if devoid of semantic information. In contrast, Grafman (see chapter 9) places highly specialized information concerned with the ordering of activity within the prefrontal cortex, in the form of structured event complexes (SEC). These provide a framework for binding together sets of actions (for example, those involved in having a meal with a friend) and also may be arranged hierarchically, with different degrees of abstraction. In this sense, the model suggests that is not just the mechanics of planning that are reliant on the prefrontal cortex, but also, in part, the material that contributes to a plan. In this fashion, prefrontal cortical damage predicts the breakdown in organization of everyday activities. Many theorists have placed executive function, including planning, in the frontal lobes, and there clearly is substantial support for this based first on studies of patients with focal lesions (see chapter 8, Morris, Bramham, & Kotitsa) and those using functioning neuroimaging (see chapter 7 by Owen). However, as a caveat, it is has been observed that many of the functions traditionally associated with the frontal cortex do not necessarily show substantive impairments with focal brain lesions. Conversely, many studies that show executive impairment with frontal lesions include patients with more widespread damage (Andres, 2003). Such observations may point towards a more distributed capacity than originally envisaged, more akin to Fuster’s (1989, 2002) model of executive functioning, which considers the cyclical interaction between anterior and posterior brain regions, or other models that incorporate the notion of more distributed capacity (D’Esposito & Grossman, 1996; Morris, 1996).

OVERVIEW OF THE CHAPTERS IN THIS VOLUME Chapters 2, 3, 4 and 5 to this volume (Davies, Ormerod, Gilhooly, & Ward) concentrate on planning in healthy young adults. All four chapters take a traditional, cognitive perspective although each focuses on different planning tasks or mechanisms. A constant theme is the dichotomy between welldefined or ill-defined problems, depending upon whether all the information necessary to solve the puzzle is presented to the participant in the instructions. If a puzzle defines the start state, the goal state, and all the rules of play (or operators) that determine the different possible moves or actions that can be taken, then a puzzle is said to be well defined.

28

WARD AND MORRIS

Davies (chapter 2) provides a general characterization of planning behaviour in well-defined problems. Central to Davies’s chapter is the issue of whether planning in such tasks is fully planned in advance or is more opportunistic in nature. He first clarifies the range of different (although sometimes interchangeable) terms that have been used to highlight this distinction: planning has been characterized as “top down” or “bottom up”, “hierarchical” or “non-hierarchical”, “total order” or “partial order”, or “goal directed” or “opportunistic”. Davies argues that these types of planning can and do occur, but little research has considered the importance of the degree of initial and concurrent planning in determining which strategy is adopted. He proposes factors determining when each type of planning is most likely to be used, and his discussion centres on issues of problem complexity, problem-solving environment and individual differences. He then examines the efficiency of initial and concurrent planning. Ormerod (chapter 3) discusses planning in ill-defined problems. In these puzzles, some of the information necessary to solve the puzzle (typically the goal state) is not fully specified. On first inspection, planning may appear to be a less than obvious option when attempting to solve ill-defined problems, because the exact representation of the goal is often not provided. Traditionally, ill-defined puzzles such as the nine-dot task and the analogous fortress/ radiation problem are assumed to require insight or some restructuring of the problem before a solution is found. However, Ormerod’s position is that planning is central to the successes and failures of finding a correct solution on this task. He details a clear planning account of these two small-scale, illdefined problems, and then extends his discussion to highly complex, realistic planning problems such as playing chess and creative design. Contrary to the claim that opportunistic planning may be more likely with ill-defined problems, Ormerod suggests that the need for structured planning increases as the problem definition decreases. The next two chapters focus their discussion on two types of mechanisms or processes that have been used to explain limitations in planning. Gilhooly (chapter 4) discusses the role of working memory in explaining planning performance. A basic premise is that our ability to temporarily hold and manipulate information in some highly accessible yet transient state(s) is limited, and that any limitation in the capacity of our working memory will constrain our planning performance. Much of Gilhooly’s chapter focuses on the five-disc TOL task. Different conceptions of working memory provide different research frameworks for planning research. Thus, Baddeley’s (1986) working memory model stimulates research using the concurrent secondary task procedure in which planning is performed with and without additional tapping, rehearsal preventing, or random generating tasks. He also introduces an alternative, individual differences approach to working memory that stimulates correlational studies in which performance on planning tasks are


29

compared with measures from other psychometric tests. Gilhooly provides evidence that working memory is not unitary but rather planning on the TOL task involves more visuo-spatial and executive processing rather than verbal short-term storage. A further indication of the limiting effects of working memory is seen in an examination of the strategies, heuristics, and verbal protocols of participants solving puzzles: they provide good evidence that memory-saving devises are indeed used. Finally, Gilhooly examines planning performance in two specialist populations – experts and older adults – whose working memory capacities are assumed to be greater in a particular planning domain or more generally limited, respectively, relative to normal young adults, and finds results consistent with a working memory perspective on planning. Ward (chapter 5) discusses the role of executive control processes when planning. Different conceptions of executive control are discussed, including the distinction made in some accounts between “lower level” and “higher level” plans and control processes. Ward summarizes four lines of evidence for the existence and control of pre-learned action plans, and then outlines a number of different attempts to study “higher level” planning. In so doing, Ward distinguishes between “planning what to do” and “planning when to do it”, but then argues that the limitations, hypothesized planning mechanisms, and research questions that arise from research of “higher order” planning are intriguingly similar to those arising from research of cognitive control of “lower level” action plans. Chapter 6 by Phillips, MacLeod, and Kliegel discusses the effects of ageing on planning ability in young and older adults. The discussion focuses on planning in the TOL task, and more open-ended planning and scheduling tasks such as the Six Elements Task (SET) and the Multiple Errands Task (MET) and their variants, both in and out of the laboratory. Phillips et al. contend that on the tower tasks, older adults are impaired in their ability to formulate and retain plans and this is caused by an age-related decline in visuo-spatial working memory. The age-related decline in planning performance in the errand tasks appears greater in abstract laboratory tasks than those framed in familiar or real-world settings, and may be based on a decline in inhibiting irrelevant information. Distinctions are also made between the formation of a plan and the execution of a plan, and the patterns of impairments seen in older healthy adults and those with frontal lobe damage. The remaining four contributions to this volume (Owen; Morris, Kotitsa, & Bramham; Grafman, Spector, & Rattermann; Burgess, Simons, Coates, & Channon) explore diverse themes relating to the cognitive neuropsychology of planning. This field has a dual purpose: attempting to determine the brain basis for planning functions, and providing a greater insight into how planning ability can be affected by brain damage. The cognitive perspectives

30

WARD AND MORRIS

or theoretical frameworks established in the first five contributions of the book are echoed throughout these chapters relating to the cognitive neuropsychology of planning. However, they also illustrate how our understanding of the cognitive aspects of planning are enhanced through considering the neurobiological underpinnings. Various themes emerge from these chapters, including the extent to which there are dedicated cognitive structures or neuronal systems that support planning activity, and the extent to which it makes sense to view planning as a psychological entity rather than an activity. The chapters illustrate the situation in which different researchers, active in the same field, support opposing frameworks in relation to these issues, suggesting the diversity of theorizing, with many issues yet to be resolved. Owen (chapter 7) reviews his research, which has accrued a substantial amount of data, principally using the TOL task either with patients or functional neuroimaging. This research provides a careful analysis of the cognitive component of solving this particular problem, including isolating the cognitive versus motoric components and considering the role of eye movements in task activity. Whilst studies of patients have provided support for prefrontal cortical involvement in TOL performance and also implicated the basal ganglia, the degree of anatomical precision of brain lesions has not been sufficient to localize such functions in any detail. Functional neuroimaging has been successful in this regard and a series of studies indicates that the dorsolateral prefrontal cortex mediates planning processes. In this chapter, the working memory demands of TOL are also considered, with Owen arriving at a similar conclusion to Gilhooly, in that there is a close interrelation between spatial working memory and planning processes, the additional aspect is that they share neuronal circuitry. Morris, Kotitsa, and Bramham (chapter 8) explore the characteristics of organization and planning impairment in patients with prefrontal cortical lesions. Their studies elucidate certain features of planning impairment in patients, including strategy formation deficits, difficulties with goal–subgoal conflict and selection equivocation. It is concluded that more work needs to be conducted to explore the development of strategies in planning tasks and their neural bases. They also outline how various techniques have been used to simulate real-world planning in the laboratory, including the development of the Virtual Planning Test (VPT) and the use of virtual reality software to bridge the gap between laboratory based methods and real-world planning. The work highlights the role of environmental context in influencing the choice of activity when activating plans and discusses how patients with frontal lobe lesions are more prone to having behaviour inappropriately “triggered” by contextual proximal activity options. Chapter 9 by Grafman, Spector and Rattermann provides an overview of different cognitive neuroscience perspectives. They reiterate the notion that


31

planning involves formulating a sequence of operations to achieve a particular goal, but that planning may involve top-down processing, controlling sequences of operations, or data-driven processes. Neuropsychological studies are taken to support the notion that plan-specific knowledge is stored primarily in the prefrontal cortex. This can be dissociated from other forms of representational knowledge, stored in posterior regions of the brain. Execution of plans may also involve motoric processes involving the basal ganglia. Hence, the studies reviewed in their chapter demonstrate the difficulties that patients with frontal cortical lesions have in developing or executing plans, but also show that those patients with subcortical lesions in regions that connect with the frontal lobes, such as the basal ganglia, also have planning impairment, albeit in a more mild fashion. A further distinction is introduced, in which there is topographical specificity for higher level planning within the prefrontal cortex, with the right cortex involved in sequencing or activity, time estimation, abstract thematic, and the left cortex involved in representing prepositional information and events. There is also a ventromedial versus dorsolateral distinction, the former involved in emotional processes to do with planning, the latter having a more “mechanistic” role. At the end of the book, a radical approach is taken by Burgess, Simons, Coates, and Channon (chapter 10), who put forward the hypothesis that there are no processes specific to planning (and consequently dedicated neuronal systems). This chapter challenges the assumption that tasks such as the TOH make the same cognitive demands as “real-world” planning. Further, the brain activity associated with such tasks may not represent planning per se, but may represent some component processes, such as generating and remembering moves. However, it is also argued that even “real-world” tasks may not be tapping specifically into “look-ahead” cognitive processes when they demonstrate impairments in patients with frontal lobe damage. Work from their own laboratory uses a set of real-life analogue tasks. By studying the decision-making processes of people when they plan, it is possible to show that many decisions are based on the semantic knowledge base of a person, which assigns priorities to certain activities, rather than “rational” problem solving. An intriguing finding from one of their studies is the lack of correlation between performance on different types of “ecologically valid” planning tasks. This could at first sight be taken to indicate low reliability or validity, but this was discounted by the presence of significant correlations with other measures, specifically non-verbal memory. This type of finding led to the conclusion that there are no “planning-specific” processes, but planning is a general term for a wide range of diverse activities.

32

WARD AND MORRIS

REFERENCES Andres, P. (2003). Frontal cortex as the central executive of working memory: Time to revise our view. Cortex, 39, 871–895. Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Andrade, J. (Ed.). (2001). Working memory in perspective. Hove, UK: Psychology Press. Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. In J. Kuhl & J. Beckermann (Eds.), Action control: From cognition to behavior (pp. 11–39). Berlin: Springer-Verlag. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human decision Processes, 50, 179–211. Anzai, Y., & Simon, H. A. (1979). The theory of learning by doing. Psychological Review, 86, 124–180. Armitage, C. J., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A meta-analytic review. British Journal of Social Psychology, 40, 471–499. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D. (1990). Human memory: Theory and practice. Hove, UK: Lawrence Erlbaum Associates Ltd. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). New York: Academic Press. Baddeley, A. D., & Logie, R. H. (1999). Working memory: The multiple-component model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 28–61). Cambridge: Cambridge University Press. Bianchi, L. (1895). The function of the frontal lobes, Brain, 18. D’Esposito, M., & Grossman, M. (1996). The physiological basis of executive function and working memory. The Neuroscientist, 2, 245–352. Friedman, S. L., & Scholnick, E. K. (Eds.). (1998a). The developmental psychology of planning: Why, how, and when do we plan? Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Friedman, S. L., & Scholnick, E. K. (1998b). An evolving “blueprint” for planning: Psychological requirements, task characteristics, and socio-cultural influences. In Friedman, S. L. & Scholnick, E. K. (Eds.), The developmental psychology of planning: Why, how, and when do we plan? Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Friedman, S. L., Scholnick, E. K., & Cocking, R. R. (Eds.). (1987). Blueprints for thinking: The role of planning in cognitive development. Cambridge: Cambridge University Press. Fuster, J. M. (1989). The prefrontal cortex: Anatomy, physiology and neuropsychology of the frontal lobe (2nd ed.). New York: Raven Press. Fuster, J. M. (2002). Physiology of executive functions: The perception–action cycle. In D. Stuss & R. T. Knight (Eds.), Principles of frontal lobe function (pp. 96–108). Oxford: Oxford University Press. Goel, V., Grafman, J., Tajik, J., Gana, S., & Danto, D. (1997). A study of the performance of patients with frontal lobe lesions in a financial planning task. Brain, 120, 1805–1822. Hayes-Roth, B., & Hayes-Roth, F. (1979). A cognitive model of planning. Cognitive Science, 3, 275–310. Hebb, D. O. (1945). Man’s frontal lobes. Archives of Neurology and Psychiatry, 54, 10–24. Hoc, J.-M. (1988). Cognitive psychology of planning. London: Academic Press. Kleist, K. (1934). Gehirnpathologie. Leipzig: Barth. Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985). Why are some problems hard? Evidence from the Tower of Hanoi. Cognitive Psychology, 17, 248–294. Lovett, M., Reder, L. M., & Lebiere, C. (1999). Modeling working memory in a unified architecture: An ACT-R perspective. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 135–182). Cambridge: Cambridge University Press.


33

Luria, A. R. (1969). The frontal syndrome. In V. J. Vinken & G. W. Bruyn (Eds.), Handbook of Clinical Neurology (Vol. 2). Amsterdam: North Holland Publishing Company. Luria, A. R. (1973). The working brain. London: Allen Lane: Penguin Press. Miller, G. A., Galanter, & Pribram, K. H. (1960). Plans and the structure of behavior. New York: Holt, Rinehart, & Winston. Miyake, A., & Shah, P. (Eds.). (1999). Models of working memory: Mechanisms of active maintenance and executive control. Cambridge: Cambridge University Press. Morris, R. G. (1996). The cognitive neuropsychology of Alzheimer-type dementia. Oxford: Oxford University Press. Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Newell, A., Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem solving. Psychological Review, 65, 151–166. Norman, D. A., & Shallice, T. (1986). Attention to action: willed and automatic control of behaviour. In R. J. Davison, G. E. Schwartz & D. Shapiro (Eds.), Consciousness and selfregulation (Vol. 4, pp. 1–18). New York: Plenum Press. Oatley, K., & Johnson-Laird, P. N. (1987). Towards a cognitive theory of emotions. Cognition & Emotion, 1, 29–50. Pavlov, I. P. (1949). Complete collected works (Vols. 1–6). Moscow and Leningrad: Izd. Akad. Nauk SSSR. (Russian) Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S. E., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology, 54A, 579–598. Pribram, K. H. (1961). A further analysis of the behavior deficit that follows injury to the primate frontal cortex. Experimental Neurology, 3, 432–466. Robbins, T. W., Anderson, E. J., Barker, D. R., Bradley, A. C., Fearnyhough, C., Henson, R., Hudson, S. R., & Baddeley, A. D. (1996). Working memory in chess. Memory and Cognition, 24, 83–93. Sacerdoti, E. D. (1974). Planning in a hierarchy of abstraction spaces. Artificial Intelligence, 5, 115–135. Sacerdoti, E. D. (1977). A structure for plans and behavior. New York: Elsevier North-Holland, Inc. Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals and understanding. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Scholnick, E. K., & Friedman, S. L. (1987). The planning construct in the psychological literature. In S. L. Friedman, E. K. Scholnick & R. R. Cocking (Eds.), Blueprints for thinking: The role of planning in cognitive development. Cambridge: Cambridge University Press. Scholnick, E. K., Friedman, S. L., & Wallner-Allen, K. E. (1998). What do they really measure? A comparative analysis of planning tasks. In S. L. Friedman & E. K. Scholnick (Eds.), The developmental psychology of planning: Why, how, and when do we plan? Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Shah, P., & Miyake, A. (1999). Models of working memory: An introduction. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 1–27). Cambridge: Cambridge University Press. Shallice, T. (1982). Specific impairments in planning. Philosophical Transactions of the Royal Society London, B298, 199–209. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T. (1994). Multiple levels of control processes. In C. Umilta & M. Moscovitch (Eds.), Attention and performance XV (pp. 395–420). Cambridge, MA: MIT Press.

34

WARD AND MORRIS

Shallice, T. (2002). Fractionation of the supervisory system. In D. Stuss & R. T. Knight (Eds.), Principles of frontal lobe function (pp. 267–277). Oxford: Oxford University Press. Shallice, T., & Burgess, P. W. (1996). The domain of supervisory processes and temporal organisation of behaviour. In A. C. Roberts, T. W. Robbins, & L. Weiskrantz (Eds.), Executive and cognitive functions of the prefrontal cortex. Philosophical Transactions of the Royal Society of London, 351, 405–412. Simon, H. A. (1975). The functional equivalence of problem-solving skills. Cognitive Psychology, 7, 268–288. Simon, H. A. (1978). Information-processing theory of human problem solving. In W. K. Estes (Ed.), Handbook of learning & cognitive processes: V. human information (pp. 271–295). Oxford: Lawrence Erlbaum Associates Ltd. Stuss, D. T., & Benson, D. F. (1987). The frontal lobes and control of cognition and memory. In E. Perecman (Ed.), The frontal lobes revisited (pp. 141–158). New York: IRBN Press. Sussman, G. J. (1975). A computer model of skill acquisition. New York: Elsevier North-Holland. Young, R. M., & Lewis, R. L. (1999). The SOAR cognitive architecture and human working memory. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 224–257). Cambridge: Cambridge University Press.

CHAPTER TWO

Planning and problem solving in well-defined domains Simon P. Davies Department of Psychology, University of Hull, UK

INTRODUCTION The ability to plan differentiates us from most other animals (Tomasello & Call, 1997). In particular, the idea that we can imagine what the future might look like and how our actions may impact upon imagined states of affairs that constitute this future must provide humans with a huge evolutionary advantage (Benson, 1993). Moreover, such abilities are implicated in larger philosophical questions including the nature of conscious experience and free will. Planning is a multidisciplinary area of research and has relevance to cognitive psychology primarily in our understanding of problem solving (e.g., Newell & Simon, 1972). Planning is also of relevance to cognitive neuroscience, where studies using planning tasks have indicated neural correlates of this activity (Fincham, Carter, van Veen, Stenger, & Anderson, 2002; Milner, Petrides, & Smith, 1985; Morris, Ahmed, Syed, & Toone, 1993; Owen, Downes, Sahakian, Polkey, & Robbins, 1990), and impairments in the ability to plan have been shown to be relevant to conditions such as autism (Hughes, Russell, & Robbins, 1994) and other generalized frontal lobe difficulties (Goel & Grafman, 1995; Goel, Pullara, & Grafman, 2001; Morris, Miotto, Feigenbaum, Bullock, & Polkey, 1997; Owen, 1997; Rattermann, Spector, Grafman, Levin, & Howard, 2001; Shallice, 1982). Planning also has relevance in other disciplines where attempts have been made to model the behaviours exhibited by humans, such as in artificial intelligence (AI), where 35

36

DAVIES

efforts have been made to explore planning behaviour from a computational perspective (Wilensky, 1983). This chapter provides a brief overview of human planning behaviour, outlining the distinctions between different forms of planning and highlighting their psychological significance in the context of research into wellstructured problems. It first considers the different forms of planning that have been identified by researchers in different research areas and argues that some aspects of the terms that are used are interchangeable. The distinction between “initial planning” and “concurrent planning” is proposed, and this distinction is argued to have some pragmatic advantages over previous terms. Next, an attempt is made to review the factors that might determine the degree of initial and concurrent planning undertaken in different planning environments. Three primary determinants of planning behaviour are identified here: problem complexity, the problem-solving environment, and inter/intra-subject differences. The chapter then considers the effectiveness of initial planning in different problem-solving situations. Inevitably perhaps, the conclusion to this chapter ends with a call for research that considers the potential interactions between these putative determinants of planning behaviour, since it seems unlikely that we will be able to fully capture the complexities of human planning by focusing upon any single factor in isolation.

CHARACTERIZING PLANNING BEHAVIOUR IN WELL-DEFINED DOMAINS Despite its widespread theoretical and practical importance, human planning remains a relatively under-explored phenomenon in cognitive psychology. Much of the research in this area has tended to focus upon problems that can be characterized as well structured or well defined. Here, participants are presented with all the information that they need to solve the problem: they are provided with the initial start state, the desired goal state, and they are told how moves can be made and all the rules of play (that is, the methods or operators) that can transform one state into another. In these well-defined puzzles, it is possible to represent the underlying structure of a problem in terms of its abstract state space which shows how all the different possible states of the puzzle are connected by the different operators. Finding a solution to such puzzles can be likened to searching the abstract state space for a pathway that connects the start state to the goal state. Rather than search the space by trial and error or by using an exhaustive algorithm, planning strategies help limit the extent of the search whilst offering a good chance of solving the puzzle successfully. For example, classical models of problem solving, such as the General

2. PLANNING AND PROBLEM SOLVING IN WELL-DEFINED DOMAINS

37

Problem Solver (Newell, Shaw, & Simon, 1958), use such representations and processes to characterize human problem solving as a systematic search of the state-space guided by general heuristics such as means–ends analysis. Here, if the solution to a problem cannot be found in a single step, the problem solver creates subgoals that may lead them closer to the solution. Such models of problem solving suggest a rational, systematic, top-down approach to planning where possible solution steps can be formulated in advance of attempts to implement the solution to the problem. Clearly, cognitive limitations, such as working memory capacity, mean that for any nontrivial problem it is not possible to plan the entire solution in a single step. Under these conditions, subgoals need to be created and plans formulated to achieve these subgoals. This rational, systematic, top-down or goal-directed approach to planning can be illustrated by considering the planning that occurs when solving a simple version of the three-disc version of the Tower of Hanoi (TOH) (see chapter 1 by Ward and Morris, Figure 1.3). Clearly, different formations of this problem (including the arrangement of discs, different cover stories and number of discs, etc.) are possible and determine the complexity of the problem. From the initial state of this problem two moves are possible – red disc to middle or red disc to right peg. Then, the black disc can be extracted, and so on. Means–ends analysis can be applied from the initial state of the problem to select a set of operators that will construct an efficient path from this state to the specified goal state. For simple versions of the TOH problem it may be possible to explore an entire sequence of moves internally. For more complex versions, subgoals need to be created that can be achieved as intermediate steps toward the solution. In terms of the above example, a subgoal from the initial state of the problem may be to extract the black disc. Means–end analysis thus describes a systematic search of the solution space following a predetermined sequence of operators from an initial to a goal state for a given problem. Clearly, problems such as the TOH are seldom encountered in everyday life, but such problems provide insight into basic planning processes whilst allowing researchers to control for other aspects of behaviour. More recent research in the problem-solving literature has attempted to characterize complex problem-solving behaviour in a rather different way. Behaviour is described as opportunistic – that is, participants respond to opportunities as they emerge rather than strictly following a predetermined plan of action (Byrne, 1977; Davies & Simplicio-Filho, 1992; Hayes-Roth & Hayes-Roth, 1979; Patalano & Seifert, 1997). For example, Hayes-Roth and Hayes-Roth (1979) demonstrated that problem solving might be described in terms of a process where the problem solvers’ current decisions and observations might suggest various unplanned opportunities for the revision of existing goals and plans. Such models suggest that planning steps are formulated

38

DAVIES

as the solution to the problem develops, rather than entirely goal directed in advance of the solution to the problem. Interestingly, similar distinctions to that between goal-directed and opportunistic planning and problem-solving behaviour have also emerged from the artificial intelligence (AI) and neuroscience literatures, but different terminology is used in these domains. For example, in the artificial intelligence (AI) literature, one distinction is between hierarchical and non-hierarchical planning (Fikes & Nilsson, 1971; Waldinger, 1977). In hierarchical planning, plans are formulated in advance of solving a problem and implemented when needed in a given order. In non-hierarchical planning, plans are implemented when and where needed and can be suspended while awaiting additional information. Similarly, in the neuroscience literature, this distinction is sometimes referred to as a contrast between partial and total order planning (Rattermann et al., 2001). It is clear that although the terminology used in different disciplines varies considerably, the meanings of these terms are potentially interchangeable. Broadly, the distinction between hierarchical/total order planning and nonhierarchical/partial order planning corresponds to the distinction made so far in this chapter between goal-directed and opportunistic planning. It is also clear that the goal-directed and the opportunistic accounts of planning and problem solving are not mutually exclusive. Hence, planning behaviour may consist sometimes of one approach, sometimes of the other and sometimes of both. However, little research has focused upon the relative efficacy of these different strategies. Given the potential overlap in terms used by researchers in different disciplines, it would be useful if it were possible to provide a general characterization of planning behaviours within a psychological context. To this end, I propose the terms initial planning and concurrent planning to help clarify the issue regarding differences in planning behaviours. The terms simply refer to the planning activity that takes place before overt problem solving action takes place (initial planning) and the planning activity that is initiated “on-line” after problem solving has commenced (concurrent planning). Clearly there is overlap between this distinction and existing nomenclatures. It is likely that solutions to problems that are initially planned (in advance) are more likely to be goal directed, hierarchical, or total order. Conversely, solutions to problems that are planned concurrently (or on-line) are more likely to be opportunistic, non-hierarchical, and partial order. However, the advantages of using the terms initial planning and concurrent planning are: (1) they are more measurable, rather than defining hypothetical cognitive activities; (2) they are theoretically neutral with respect to what is actually happening during this time; (3) they are defined as umbrella terms providing a catch-all for the existing nomenclatures.


39

SELECTION AND EFFECTIVENESS OF DIFFERENT PLANNING BEHAVIOURS There are contrasting patterns of advantages and disadvantages incurred when undertaking initial and concurrent planning. One advantage of concurrent planning is that it may reduce the load on working memory since the current state of the problem is encoded by the problem display. However, it also has associated difficulties since it may be hard to backtrack having made a wrong move. By contrast, initial planning clearly imposes a high working memory load, but the flexibility of storage in working memory means that different alternatives can be considered at relatively low cost. However, the debate between those who advocate systematic methods and heuristics, such as means–ends analysis, compared with those who advocate unstructured, opportunistic planning behaviour rarely considers the relative importance of concurrent and initial planning, except rather indirectly, and several specific important empirical questions concerning the relationship between planning and performance remain unexplored. The first question is what factors determine the degree of initial and concurrent planning behaviour when problem solving. In the next three sections, three factors are identified that may affect the degree of initial and concurrent planning when problem solving. These factors are (1) the complexity of the problem; (2) the problem environment; (3) individual and group differences. A second question is whether planning some or all of the solutions to a problem before engaging in problem-solving behaviour leads to better or worse performance on a particular problem, compared with planning a solution concurrently with the implementation of that solution. The effectiveness of initial planning is considered in a fourth and final section.

PROBLEM COMPLEXITY AND PLANNING STRATEGY The idea that planning occurs completely in advance may only be accurate for problems exhibiting a certain limited level of complexity. Consider for example the TOH problem. For three-ring versions of this problem it is quite feasible to plan the entire solution to the problem internally. For four-ring versions it is possible to plan most of the solution, but at greater levels of complexity; initial planning does not intuitively seem productive, particularly if one considers the rather volatile and error-prone nature of working memory. As Karat (1982) has pointed out, participants typically do not have a complete understanding of the TOH before they begin to implement their solution. More recent empirical research (Phillips, Wynn, Gilhooly, Della Sala, & Logie, 1999; Phillips, Wynn, McPherson, & Gilhooly, 2001) has suggested that prior planning actually has little effect on performance for

40

DAVIES

moderately complex versions of the TOL task (see chapter 6 by Phillips, Macleod, & Kliegal, Figure 6.1). However, it is difficult to establish comparable measures of complexity when comparing different problems and it may well be the case that the effectiveness of prior planning is moderated by problem complexity. Few studies have actually considered the effects of initial planning on solutions to well-structured problems. One exception to this is a recent study by Ward and Allport (1997), who considered the relationships between amount of preparation time and problem-solving performance. In this study, the participants were given an open-ended preparation time phase and then had to solve the puzzles as quickly as possible. Ward and Allport found that the time taken to prepare a solution increased monotonically with problem complexity. Ward and Allport also reported that little additional planning took place during problem solving, and that there was also a negative correlation between participants’ error and planning times. However, the finding that planning time increases with problem complexity is not the same thing as showing that participants were always planning fully in advance during this initial time. On the one hand, participants may have been performing initial planning only on the most simple puzzles. Indeed, the finding of a negative correlation between error and planning time may reflect performance on only a subset of puzzles. Alternatively, because they used the TOL task, which lacks the size constraint of the standard TOH in that smaller rings can only be stacked on top of larger rings, it is relatively easy to perform legal moves that help attain the goal state, and there is relatively little disincentive if the wrong move is made. It could be that participants find enough correct solutions through concurrent planning and this may discourage some participants to perform any more effortful initial planning. One might argue that the effects of an initial preparation phase may well disappear with more complex problems (as shown in the Phillips et al. study, 2001). Hence, the findings of the Ward and Allport study may not be generalizable to the most complex problems that they employed and it may be impossible to mentally simulate more than a small number of potential moves. The idea that initial planning may only be effective for moderately difficult puzzles is supported by a study (to be discussed more fully later) by Davies (2003) who investigated the effects of initial planning time on puzzles of a range of difficulties. While initial planning may be of some benefit to problemsolving performance, it appears to be restricted to problems of a moderate complexity level and did not confer any particular advantage to those solving more difficult problems.


41

PROBLEM-SOLVING ENVIRONMENT AND PLANNING STRATEGY There are many different ways in which the problem-solving environment can help encourage or discourage the use of an initial planning strategy over a concurrent planning strategy. Three examples are: (a) the effects of the problem-solving display; (b) the effects of changes in the ease of use of the problem interface; (c) the effects of changes in the task instructions, particularly the instruction to verbalize during problem solving. Research on opportunistic planning has typically been restricted to illstructured problem-solving domains, but it may also have parallels in domains that display greater formal structure. Several researchers have investigated the idea that changes to the external features of certain isomorphic problems (problems with the same abstract problem space but with a different cover story) may affect the ease with which such problems can be solved (Kotovsky, Hayes, & Simon, 1985). Initial interest in the role played by problem-solving environments maintained a primary role for cognition in problem-solving behaviour. However, more recent accounts of such behaviour have begun to emphasize the central role played by the external environment in the mediation of cognitive activity. For example, theories of display-based problem solving (Davies, 1992; Larkin, 1989; Mannes & Kintsch, 1991; Zhang, 1997) seek to explain certain facets of behaviour by stressing the contribution of external memory sources which act as repositories for search control knowledge, and intermediate state information. The main idea promoted by such models is that information contained within a display can reduce the complexity of the mental processes required to perform certain problem-solving tasks. Hence, display-based strategies can partially supplant a complex plan/goal structure by enabling problem solvers to substitute efficient perceptual operations for typically unreliable cognitive processes. In general, these models suggest that the planning processes involved in problem solving are likely to be task concurrent. In some extreme cases the role of planning is minimized since solution steps may be perceptually obvious to the solver, but typically some role for planning is admitted by such models. Recent work suggests that one can encourage different planning strategies by affecting the ability to implement operators and this can be taken as evidence that the characteristics of the problem environment may affect which of the two different planning strategies participants adopt. O’Hara and Payne (1998) report a study where they compared performance using two different computer interfaces to a problem-solving task. In one condition, participants were requested to manipulate the aspects of the problem using a mouse (a form of direct manipulation). In a second condition, the same problem was manipulated via text-based instructions that were typed by the participants. In a subsequent post-experimental test, O’Hara and Payne

42

DAVIES

found that those participants who were asked to type instructions not only performed better than the direct manipulation group, but also demonstrated more evidence of planning behaviour when they provided verbal protocols. Thus, by increasing the cost of making additional moves, the effectiveness of initial planning was increased. A third way in which the problem-solving environment may affect the strategies that are employed is through the instructions given to participants. For example, one can ask participants to focus upon speed or accuracy and this manipulation in instruction appears to have significant effects upon the strategies that they use (Fum & Del Missier, 2001; Haider & Frensch, 1999; Unterrainer, Rahm, Leonhart, Ruff, & Halsband, 2003). Instructions can also be manipulated by asking participants to evaluate their problem-solving performance (Davies, 2000) or to verbalize whilst solving a problem (AhlumHeath & Di Vesta, 1986). It is now widely accepted that verbalization whilst solving a problem can significantly improve problem-solving performance (Ahlum-Heath & Di Vesta, 1986; Berardi-Coletta, Buyer, Dominowski, & Rellinger, 1995; Berry, 1983; Davies 2000; Gagne & Smith, 1962; McGeorge & Burton, 1989). However, there is some debate in the literature about the role played by verbalization per se and the role played by evaluation. For example, it has been argued that the act of verbalizing encourages participants to focus upon the evaluation of existing and future problem states. Moreover, if participants are instructed to think about how moves might be evaluated without verbalizing, their performance is indistinguishable from those who verbalize about the same processes (Berardi-Coletta et al., 1995). However, one of the main conclusions to emerge from such studies is that verbalization may encourage participants to move away from an exploratory, opportunistic mode of problem solving to a more goal-directed and planful approach. Questions that remain unanswered include why a particular mode of planning may not be spontaneous (especially when the nonpreferred mode of planning may confer a performance advantage), and why strategies can be affected significantly by the mode of instruction given to participants. It should be noted that other interpretations of the changes in performance seen to be associated with verbalization may be suggested. One possibility is that giving verbalizations may in some way cause strategic changes (Davies & Castell, 1992) that can be from initial to concurrent planning as well as from concurrent to initial planning. For, example, Davies (1995) has argued that the elicitation of concurrent verbal protocols during a design task may in fact cause opportunistic deviations from a top-down approach. This is based upon the notion that opportunistic episodes arise from simple cognitive failures where information is lost from working memory (Anderson, 1983; Anderson & Jeffries, 1985). This suggests that those points in a design


43

task that are cognitively taxing will be made more difficult when a designer has to articulate their problem-solving behaviour. Moreover, the use of verbal protocols to characterize temporally bound events, such as the kind of problem solving that occurs in design decomposition, may have serious implications for the way in which such events are described. Davies (1995) reports an empirical study of a software design task which demonstrates that described behaviour elicited in the form of concurrent verbal protocols can differ significantly from observed behaviour. In particular, this study showed that designers tend to articulate goal structures in a particular temporal sequence when describing their activity, whereas the implementation of the design follows a different temporal sequence. It could be argued that this arises because of a tendency to linearize verbal descriptions of non-linear behaviour. Language use is clearly structured and temporally sequenced (Levelt, 1981) and this structure may well not mesh with descriptions of processes that have a non-predictable and arbitrary structure. This suggests that some care should be taken when using a methodology such as protocol analysis to characterize events that have a significant temporal or process-based dimension such as software design, and probably other problem-solving processes. To some extent these problems are not new. Ericsson and Simon (1980) suggested that participants may stop verbalizing in conditions of high cognitive load. We might also expect the reverse to be the case. Hence, verbalization may affect the process participants are trying to describe, especially in complex situations where working memory and short-term memory are taxed. Although this argument remains a possibility, there is sometimes little effect of verbalization on performance. Thus, Newell and Simon (1972) looked at the behaviour of participants solving propositional logic problems in a condition in which participants concurrently verbalized their problemsolving behaviour and compared it with a condition where participants were not required to provide protocols. They found no significant performance differences in terms of the deployment of correct solution steps. However, with the above exception reported by Davies (1995), I am unaware of any studies that have looked more directly at the effects of verbalization upon strategy in complex problem-solving tasks. Another perhaps related reason for the differences between described and actual behaviour in design tasks is based upon the idea that rational analysis is used to explicate goals and structure, whereas actual design behaviour is possibly based upon the recognition of opportunities and is strongly data driven (Davies & Simplicio-Filho, 1992). The suggestion here is that goaloriented behaviour stems from the level of bounded rationality (Newell & Card, 1985), whereas opportunism stems from lower psychological levels that may not be amenable to verbalization. This distinction is made more explicit in other cognitive architectures. For example, in SOAR (Laird, Newell, &

44

DAVIES

Rosenbloom, 1987) rational action related to knowledge and goals comprised the knowledge level of human cognitive activity (Newell, 1982). This is underpinned by lower levels of cognitive behaviour, such as recognition memory and automatic decision cycles that retrieve information from longterm memory. We might characterize opportunistic behaviour as being driven by these lower level psychological mechanisms. In contrast, the expression of goals and rational action will be based upon other factors and will be determined to a large extent by the techniques and practices employed by participants in these studies. In summary, it is clear that any characterization of planning processes in problem solving must take account of the environment in which problem solving occurs. For example, some environments are likely to support opportunistic problem-solving behaviour, especially where goals and subgoals can be easily externalized, thus reducing the potential load on working memory. The ability to return to this externalized information to make changes also seems to be a prerequisite to opportunistic planning. Moreover, environments that display the flexibility to support such a process are also likely to facilitate opportunistic planning processes. One might also argue that some care might be exercised in the interpretation of reported behaviours since there may be some reason to suspect that the act of verbalizing may actually alter the behaviours that are being reported or that verbalizing may structure and rationalize such behaviours.

INDIVIDUAL AND GROUP DIFFERENCES AND PLANNING STRATEGY So far we have considered the role played by problem complexity and the problem-solving environment in planning behaviour. Another important issue relates to individual preference for a particular approach to planning. A number of studies have considered individual differences in the determination of problem-solving strategy (Haygood & Johnson, 1983; MacLeod, Hunt, & Mathews, 1978; Roberts, Gilmore, & Woods, 1997; Sternberg, 1977). For example, Sternberg (1977) found that better reasoners spent more time in an initial encoding phase than those who were less proficient. Roberts et al. (1997) demonstrated systematic differences in strategy preference, which were related to spatial reasoning ability. Furthermore, Hegarty and Sims (1994) showed that spatial visualization ability was correlated with performance on a mechanical reasoning task requiring mental animation, and presumably some degree of forward planning. The choice of planning strategy adopted by a participant is likely to be affected by their experience at solving the problem. For example, Anzai and Simon (1979) demonstrated changes in strategy as problem solvers become more proficient with the TOH problem. In general terms, Anzai and Simon


45

showed that during inexperienced phases of problem solving, participants tend to exhibit behaviour that might be characterized as exploring the problem space. As experience increases, more systematic methods are used such as means–ends analysis as participants learn better sequences of moves and rule out bad moves. Moreover, within a given level of experience, problem solvers have been shown to adopt a range of potential strategies. Simon (1975) identifies several strategies for solving the TOH, including goal recursion and perceptual strategies. Perceptual strategies use current perceptual features of a problem (such as certain spatial configurations of discs) to determine the next move, whereas goal recursion strategies involve maintaining a stack of goals in working memory. There may be parallels here between these strategies and those identified in this chapter. In particular, the goal recursion strategy relies upon working memory to a much greater degree than the perceptual strategy and may well demand a good deal of planning activity. Perceptual strategies, on the other hand, may involve planning which tend to be task concurrent. There is also some evidence in the recent literature for systematic differences in the planning exhibited by different groups of participants. Rattermann et al. (2001) used two groups of children (aged 7 to 8 and 11 to 13) and two groups of adults (a control group and a group with prefrontal cortex damage). Rattermann et al. were interested in total and partial order planning – terms derived from artificial intelligence (AI) models of planning which broadly correspond to the distinction between initial and concurrent planning that has been presented here. They found that the 7- to 8-year-old children and prefrontal cortex damaged patients’ behaviour corresponded to total planning, whereas the 11- to 13-year-old children and adult controls exhibited behaviour corresponding to partial order planning which, in turn, has been shown to be more efficient in computationally based planning systems. These findings may suggest that total order planning is a suboptimal strategy that has both developmental precursors and correlates with clinical populations. Other studies have considered whether age may affect planning processes in adult participants. Gilhooly, Phillips, Wynn, Logie, and Della Sala (1999) report a study of planning processes for the five-disc TOL task performed by 20 younger and 20 older adult participants. Gilhooly et al. used a concurrent “think-aloud” method to obtain data on planning processes prior to moving discs in the TOL. The main finding of this study was that older and younger participants did not differ in average moves taken to solve the tasks. However, older participants’ planning was less complete and more error prone than that of younger participants. Age differences were found in an initial planning stage, during which there was no stimulus support and presumably a substantial working memory load. During the move phase there was stimulus support and hence little loading of working memory. Age differences in

46

DAVIES

moves required were not found in the move phase. Gilhooly et al. argued that older participants tended to have depleted working memory capacities and that working memory is heavily loaded in TOL planning but less heavily loaded in the move phase of the task. Other group differences emerge in studies of the way in which strategies may change with developing expertise or in terms of the differences between novices and experts in particular areas such as physics (Larkin, McDermott, Simon, & Simon, 1980), chess (Holding & Reynolds, 1982) and computer programming (Davies, 1996; Rist, 1995). For example, Davies (1996) looked at display-based planning and expertise in the context of a computer programming task. This study showed that experts relied upon external sources to record program fragments as these were generated and then returned later, in terms of the temporal sequence of program generation, to further elaborate these fragments. It has been suggested that a major determinant of expertise in programming may be related to the adoption or development of strategies that facilitate the efficient use of external sources (Davies, 1993). The externalization of information clearly has a high cost in terms of the reparsing or re-comprehension of generated code that is implied. Hence, it might seem counter-intuitive to suggest that problem solvers will tend to rely upon this kind of strategy rather than upon a strategy that involves the more extensive use of working memory. However, this explanation is consonant with existing work that has implicated display-based recognition skills in theoretical analyses of complex problem solving (Larkin, 1989). The contribution of these analyses has been important, but they have neglected to consider the relationship between display use and expertise and the consequent effect that this may have upon the nature of problem-solving strategies.

EFFECTIVENESS OF INITIAL PLANNING This final section addresses the effectiveness of initial planning. As we have seen, the adoption of a particular strategy is likely to be dependent upon the complexity of a problem, as well as the environment of the problem, and individual differences. For example, it has been argued that more complex problems demand a certain degree of concurrent planning and one specific question is whether initial planning actually makes any contribution to the solution of such problems. This question may be difficult to answer from the current problem-solving literature because many studies reported in the literature give participants little or no preparation time and this may encourage, or indeed force participants to adopt a concurrent planning strategy. In many cases, participants are allowed to read the experimental instructions and then begin to solve the problem. In other studies, preparation time is self-paced and this can be


47

problematic in terms of introducing potential confounds, particularly if individuals differ systematically in their preference for a particular problemsolving strategy. As has been pointed out, these two strategies are clearly not mutually exclusive but an analysis of the relative contribution of each will clearly contribute to a theoretical understanding of problem-solving behaviour. One set of relevant data to this issue is that reported by Davies (2003), which was alluded to earlier. In this study, groups of participants were initially screened to determine the amount of time they had spent in an openended preparation phase before solving a TOH problem. One significant finding of this study was that while initial planning may be of some benefit to problem-solving performance, it appears to be restricted to problems of a moderate complexity level and did not confer any particular advantage to those solving more difficult problems. Moreover, there appeared to be strong individual preferences for particular strategies, which may in turn be related to individual differences in such things as spatial reasoning or forward planning abilities. Whether preferences for such strategies are related to individual differences and what the nature of these differences might be clearly remains an open empirical question, and more research will be needed to determine whether this preference is dependent upon spatial visualization or some substrate of this related to more fundamental components of working memory, for example, central executive processes. It appears that both the ability to simulate future moves, combined with central executive processes, may work in tandem to determine planning strategies that are adopted in solutions to the kinds of problems studied by Davies (2003). However, evidence for two distinct modes of planning seems clear at least in the context of problems of a given level of complexity and of the kind reported by Davies (2003). Whether or not these distinct forms of planning strategy are applicable to other kinds of problems suggests another avenue of research. Clearly, well-structured problems have fairly unique characteristics that may not be shared by other categories of problem. Where the effects of initial planning have been studied, it has typically been in the context of ill-structured problems where external support for working memory has been available, for example, in the area of text composition (Rau & Sebrechts, 1996). Clearly, using well-structured problems is advantageous in terms of being able to control for various confounding variables and in relation to the systematic measures that one can use to assess performance under such circumstances. Davies highlights strong differences in the planning behaviours of participants solving a restricted class of well-structured problems. The ability to draw general conclusions about planning and problem-solving behaviour from studies of TOH problems to other problems is clearly restricted, however,

48

DAVIES

and it remains to be seen whether these planning differences generalize to other kinds of problems.

SUMMARY AND CONCLUSIONS This chapter has reviewed the growing literature on planning in the context of well-structured problem solving. A primary distinction between initial and concurrent (opportunistic) planning has been highlighted and possible determinants of these different planning behaviours have been discussed. The literature suggests three factors that may determine planning behaviour. First, we considered the complexity of the problem itself. It seems clear that when problems remain relatively simple much of the solution to the problem can be developed in working memory. It is possible that more complex problems necessitate some degree of concurrent planning and when there is no possibility of externalizing the contents of working memory, pre-planning a solution may turn out to be counterproductive given the rather volatile and error prone nature of storage in working memory. Second, some environments are likely to support opportunistic problem-solving behaviour, especially where goals and subgoals can be easily externalized, thus reducing the potential load on working memory. The ability to return to this externalized information to make changes also seems to be a prerequisite to opportunistic planning. Third, systematic inter- and intra-group differences may exist in approaches to planning. For example, it is fairly widely reported that novices and experts in a particular domain may exhibit different strategies. Moreover, even within a single problem-solving episode planning strategies may change. Less emphasis has been placed on the possibility that there may be individual differences in preferred planning strategy, but some evidence is now emerging to support such a view. Finally, the effectiveness of initial planning has typically been ignored (or has not been controlled) in many studies of problem solving, but this phase is clearly an important stage in the problem-solving process, especially of puzzles of moderate difficulty. It may be a major factor in determining the strategies used whilst solving a problem. Planning is clearly a complex activity and research should consider the potential interactions between the putative determinants of planning behaviour that have been identified in this chapter. It seems unlikely that we will be able to fully capture the complexities of human planning by focusing upon any single factor in isolation.

REFERENCES Ahlum-Heath, M. E., & Di Vesta, F. J. (1986). The effect of conscious controlled verbalization of a cognitive strategy on transfer in problem solving. Memory and Cognition, 14, 281–285.


49

Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press. Anderson, J. R., & Jeffries, R. (1985). Novice LISP errors: Undetected losses from working memory. Human–Computer Interaction, 1, 107–131. Anzai, Y., & Simon, H. A. (1979). The theory of learning by doing. Psychological Review, 86, 124–140. Benson, D. F. (1993). Prefrontal abilities. Behavioural Neurology, 6, 75–81. Berardi-Coletta, B., Buyer, L. S., Dominowski, R. L., & Rellinger, E. R. (1995). Metacognition and problem solving: A process-oriented approach. Journal of Experimental Psychology: Learning, Memory and Cognition, 21, 1, 205–223. Berry, D. C. (1983). Metacognitive experience and transfer of logical reasoning. Quarterly Journal of Experimental Psychology, 35A, 39–49. Byrne, R. W. (1977). Planning meals: Problem solving on a real data-base. Cognition, 5, 287–332. Davies, S. P. (1991). The role of notation and knowledge representation in the determination of programming strategy: A framework for integrating models of programming behaviour. Cognitive Science, 15, 547–572. Davies, S. P. (1992). The role of expertise in the development of display-based problem solving strategies. Proceedings of the 14th Annual Conference of the Cognitive Science Society (pp. 797–802). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Davies, S. P. (1993). Externalising information during coding activities: Effects of expertise, environment and task. Proceedings of the 5th Empirical Studies of Programmers Workshop, Palo Alto, USA. Norwood, NJ: Ablex. Davies, S. P. (1995). Effects of concurrent verbalisation on design problem solving. Design Studies, 16, 102–116. Davies, S. P. (1996). Display-based problem solving strategies in computer programming. In W. D. Gray & D. A. Boehm-Davis (Eds.), Empirical Studies of Programmers: Sixth Workshop. Ablex, NJ: Norwood. Davies, S. P. (2000). Move evaluation as a predictor and moderator of success in solutions to well structured problems. Quarterly Journal of Experimental Psychology, 53A, 4, 1186–1201. Davies, S. P. (2003). Initial and concurrent planning in solutions to well structured problems. Quarterly Journal of Experimental Psychology, 56A, 7, 1147–1164. Davies, S. P., & Castell, A. M. (1992). Contextualizing design: Narratives and rationalisation in empirical studies of software design. Design Studies, 13, 379–392. Davies, S. P., & Simplicio-Filho, F. (1992). Opportunistic and goal-oriented behaviour in software design: Combining empirical and theoretical studies in cognitive modelling. In J. S. Gero (Ed.), Artificial intelligence in design (pp. 839–860). Dordrecht: Kluwer Academic Press. Ericsson, K. A., & Simon, H. A. (1980). Verbal reports as data. Psychological Review, 87, 215–251. Fikes, R. E., & Nilsson, N. J. (1971). STRIPS: A new approach to the application of theorom proving to problem solving. Artificial Intelligence, 2, 189–208. Fincham, J. M., Carter, C. S., van Veen, V., Stenger, V. A., & Anderson, J. R. (2002). Neural mechanisms underlying planning. A computational analysis using event related fMRI. Proceedings of the National Acadamy of Sciences USA, 99, 3346–3351. Flower, L., & Hayes, J. (1980). The dynamics of composing: Making plans and juggling constraints. In L. Gregg and E. Steinberg (Eds.), Cognitive processes in writing. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Fum, D., & Del Missier, F. (2001). Adaptive selection of problem solving strategies. In Proceedings of the Twenty-Third Annual Conference of the Cognitive Science Society (pp. 313–318). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Gagne, R. M., & Smith, E. C. (1962). A study of the effects of verbalisation on problem solving. Journal of Experimental Psychology, 63, 12–18. Gilhooly, K. J., Phillips, L. H., Wynn, V. E., Logie, R. H., & Della Sala, J. (1999). Planning processes and age in the five-disk Tower of London task. Thinking and Reasoning, 5, 339–361.

50

DAVIES

Goel, V., & Grafman, J. (1995). Are the frontal lobes implicated in planning function: Interpreting data from the Tower of Hanoi. Neuropsychologia, 33, 623–642. Goel, V., Pullara, S. D., & Grafman, J. (2001). A computational model of frontal lobe dysfunction: Working memory and the Tower of Hanoi task. Cognitive Science, 25, 2. Haider, H., & Frensch, P. A. (1999). Information reduction during skill acquisition: The influence of task instruction. Journal of Experimental Psychology: Applied, 5, 129–151. Hayes-Roth, B., & Hayes-Roth, F. (1979). A cognitive model of planning. Cognitive Science, 3, 275–310. Haygood, R. C., & Johnson, D. F. (1983). Focus-shift and individual differences in the Sternberg memory search task. Acta Psychologica, 53, 129–139. Hegarty, M., & Sims, V. K. (1994). Individual differences in mental reasoning animation during mechanical reasoning. Memory and Cognition, 22, 411–430. Holding, D. H., & Reynolds, J. R. (1982). Recall or evaluation of chess positions as determinants of chess skill. Memory and Cognition, 10, 237–242. Hughes, C., Russell, J., & Robbins, T. W. (1994). Evidence for executive disfunction in autism. Neuropsychologia, 32, 477–492. Karat, J. (1982). A model of problem-solving with incomplete constraint knowledge. Cognitive Psychology, 14, 538–559. Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985). Why are some problems hard? Evidence from the Tower of Hanoi. Cognitive Psychology, 17, 248–294. Laird, J. E., Newell, A., & Rosenbloom, P. S. (1987). SOAR: An architecture for general intelligence. Artificial Intelligence, 18, 1–64. Larkin, J. H. (1989). Display-based problem solving. In D. Klahr and K. Kotovsky (Eds.), Complex information processing; The impact of Herbert A. Simon. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Larkin, J. H., McDermott, J., Simon, D., & Simon H. A. (1980). Expert and novice performance in solving physics problems. Science, 208, 1335–1342. Levelt, W. J. M. (1981). The speaker’s linearisation problem. Phil. Trans. Royal Society, 295, 305–315. McGeorge, P., & Burton, A. M. (1989). The effects of concurrent verbalisation on performance in a dynamic systems task. British Journal of Psychology, 80, 455–465. MacLeod, C. M., Hunt, E. B., & Mathews, N. N., (1978). Individual differences in the verification of sentence–picture relationships. Journal of Verbal Learning and Verbal Behavior, 17, 493–507. Mannes, S. M., & Kintsch, W. (1991). Routine computing tasks: Planning as understanding. Cognitive Science, 15, 305–342. Milner, B., Petrides, M., & Smith, M. L. (1985). Frontal lobes and the temporal organisation of memory. Human Neurobiology, 4, 137–142. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, B. K. (1993). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367–1378. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., & Polkey, C. E. (1997). The effects of subgoal conflict on planning ability after frontal lobe lessions in man. Neuropsychologia, 35, 1147–1157. Newell, A. (1982). The knowledge level. Artificial Intelligence, 18, 87–127. Newell, A., & Card, S. (1985). The prospects for psychological science in human–computer interaction. Human–Computer Interaction, 1, 209–242. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Newell, A. Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem solving. Psychological Review, 65, 151–166. O’Hara, K., & Payne, S. J. (1998). The effects of operator implementation costs on planfulness of problem solving and learning. Cognitive Psychology, 35, 34–70.


51

Owen, A. M. (1997). Cognitive planning in humans: Neuropsychological, neuroanatomical and neuropharmacological perspectives. Progress in Neurobiology, 53, 431–450. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1990). Planning and spatial working memory following frontal lobe lesions in man. Neuropsychologia, 28, 1021–1034. Patalano, A. L., & Seifert, C. M. (1997). Opportunistic planning: Being reminded of pending goals. Cognitive Psychology, 34, 1–36. Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S. E., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology, 54, 579–597. Rattermann, M. J., Spector, L., Grafman, J., Levin, H., & Howard, H. (2001). Partial and total order planning: Evidence from normal and prefrontally damaged populations. Cognitive Science, 25, 941–975. Rau, P. S., & Sebrechts, M. M. (1996). How initial plans mediate the expansion and resolution of options in writing. Quarterly Journal of Experimental Psychology, 49A, 616–638. Rist, R. S. (1995). Program structure and design. Cognitive Science, 19, 507–562. Roberts, M. J., Gilmore, D. J., & Woods, D. J. (1997). Individual differences in strategy selection in reasoning. British Journal of Psychology, 88, 473–492. Shallice, T. (1982). Specific imparements of planning. Philosophical Transactions of the Royal Society London, 298, 199–209. Simon. H. A. (1975). The functional equivalence of problem solving skills. Cognitive Psychology, 7, 268–288. Sternberg, R. H. (1977). Intelligence, information processing and analogical reasoning: The componential analysis of human abilities. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Tomasello, M., & Call, J. (1997). Primate cognition. Oxford: Oxford University Press. Unterrainer, J. M., Rahm, B., Leonhart, R., Ruff, C. C., & Halsband, U. (2003). The Tower of London: the impact of instructions, cueing, and learning on planning abilities. Cognitive Brain Research, 17, 675–683. Waldinger, R. (1977). Achieving several goals simultaneously. In E. W. Elcoak and D. Michie (Eds.), Machine intelligence 8. New York: Wiley. Ward, G., & Allport, D. A. (1997). Planning and problem-solving using the five disk Tower of London. Quarterly Journal of Experimental Psychology, 50A, 49–78. Wilensky, R. (1983). Planning and understanding. Reading, MA: Addison-Wesley. Zhang, J. (1997). The nature of external representation in problem solving. Cognitive Science, 21, 179–218.

CHAPTER THREE

Planning and ill-defined problems Thomas C. Ormerod Psychology Department, Lancaster University, UK

INTRODUCTION The distinction between well-defined and ill-defined problems has its origins in the specification of components of a problem space (see Hayes, 1978), that is, the space of possible move sequences given the context in which the problem is set and the information-processing limitations of the problem solver. Solving well-defined problems (as defined by Davies, chapter 2, this volume) can be viewed as a task of navigating from the start state to the goal state by applying operators at appropriate times to shift from one problem state to another under given constraints. Planning involves the evaluation of moves (i.e., choices of operators and the ways in which they are applied) in advance of their selection in an attempt to discover a sequence of one or more moves that optimizes the route from start state to goal state. Heuristics, such as hill climbing and means–ends analysis, are central to explanations of human performance with well-defined problems, such as the Tower of Hanoi (TOH) and Missionaries and Cannibals puzzles (e.g., Anderson, 1993; Simon & Reed, 1976). Under these heuristics, the problem solver plans moves by assessing the difference between the current state and the goal state and then selecting or establishing sequences of operators that minimize this difference. Ill-defined problems present a dilemma for planning: how can one plan the route towards a solution if one knows so little about the path ahead, especially when one does not know the final destination or goal state. Puzzles such as the nine-dot and radiation problems (illustrated in Figure 3.1 and Box 3.1, 53

54

ORMEROD

Figure 3.1 The nine-dot problem (a) and its solution (b). The task is to cancel each of the dots by drawing four straight lines without retracing a line or lifting the pencil from the paper. The figure labelled (c) shows a typical attempt made by a participant planning moves at two-lookahead.

Box 3.1

Duncker’s (1945) radiation problem

Suppose you are a doctor faced with a patient who has a malignant tumour in his or her stomach. It is impossible to operate on the patient, but unless the tumour is destroyed the patient will die. There is a special type of ray that can be used to destroy the tumour, as long as the rays reach the tumour with sufficient intensity. However, at the necessary intensity, the healthy tissue that the rays pass through will also be destroyed and the patient will die. At lower intensities, the rays are harmless but they will not affect the tumour either. What procedure might the doctor employ to destroy the tumour with the rays, at the same time avoiding destroying any healthy tissue?

respectively) are considered ill-defined because components of their problem spaces are not fully specified. In particular, the problem descriptions lack a statement of a concrete and visualizable goal state. Heuristics that evaluate the progress made towards a goal state are not obviously applicable to solving ill-defined problems, because it is difficult to describe a test for the final state that could be used in evaluating progress (VanLehn, 1989). However, faced with an ill-defined problem, individuals can and do plan. In fact, planning is central to both failure and success at solving ill-defined problems. Moreover, planning lies at the heart of commonalities among expert problem solvers. In this chapter, we distinguish between local planning, where choices are made between alternative moves from any particular problem state, and global planning, where decisions are made about how to structure problemsolving activity as a whole. We also distinguish between the application of pre-compiled plans retrieved from prior knowledge, and on-line planning consisting of the assembly and evaluation of sequences of moves in real time (where moves comprise actions and/or ideas that the problem solver generates

3. PLANNING AND ILL-DEFINED PROBLEMS

55

as discrete attempts to make progress towards solution discovery), either at the commencement of problem solving or during the execution of problem moves. We begin by focusing upon small-scale, ill-defined puzzles, before turning our attention to large-scale and realistic problem-solving activities.

PLANNING IN PUZZLE SOLVING The nine-dot problem (Figure 3.1) exemplifies a small-scale, ill-defined problem. Its start state is clear – an array of unconnected dots – as is the only available operator – to draw straight lines. The constraints upon what moves may be legally made by applying the operator are, in principle, known in advance. However, the goal state is under-specified: indeed, if the goal state were known in advance, there would be no problem to solve. As ill-defined problems go, the nine-dot problem is reasonably well defined. Arguably, Duncker’s (1945) radiation problem (Box 3.1) is much less well defined. The start state is specified but difficult to visualize, the operators that might be applied are apparently limitless, some constraints are present in the problem statement but are embedded in a complex description while others depend upon prior knowledge and beliefs, and the goal state is described only in abstract terms. As we shall see, the planning activities that people undertake in attempting to solve the nine-dot and radiation problems explain why they are so difficult (typically fewer than 10% of people solve either problem within a 10-minute period). However, it is only through planning that the problems can be solved at all.

Using prior problems as solution plans In the absence of concrete and visualizable goal states, it has generally been assumed that individuals do not use hill climbing or means–ends analysis, but instead use alternative heuristics to guide move selection that do not require complete goal information. The heuristic on which the vast majority of the literature on ill-defined problems focuses is structural analogy (e.g., Anderson, 1993; Gick & Holyoak, 1980). Although not usually conceived of in such terms, analogical problem solving can be characterized as a plan-based heuristic. In drawing analogies, one effectively uses the solution to a source problem as a plan for solving a target problem. The analogical re-use of pre-compiled plans is related to the concept of total order planning (see Davies, chapter 2 this volume), but differs in being based upon the retrieval of existing plans rather than the construction of new plans prior to commencing move attempts. Gick and Holyoak (1980, 1983) demonstrated the potential for analogical problem solving as a heuristic for solving ill-defined problems. They presented participants with source problems that had the same underlying solution

56

ORMEROD

concept (splitting and convergence) as Duncker’s radiation problem but different superficial descriptions (e.g., a story about a general trying to attack a well-defended fortress). In a number of experiments, they demonstrated that solutions to the radiation problem could be facilitated by presenting analogical source problems and solutions. Their findings led them to conclude that analogy lies at the core of human learning. They proposed that, through multiple analogical problem-solving episodes, individuals induce abstract schematic representations of problems and their solutions can be retrieved and applied as solution plans when structurally similar problems are presented. In a similar vein, Anderson (1993) has suggested that structural analogy is the key heuristic that leads to the compilation of new procedures for action. More recently, Thompson, Gentner, and Lowenstein (2000) have argued that the same kinds of analogical heuristic mediate skilled problem solving in practical domains such as managerial negotiation. In almost all the studies that demonstrate successful analogical transfer of a conceptual solution to a superficially different target problem, participants are either given a hint that the target and source problems are related, or are otherwise placed in situations where the requirement to analogize is unavoidable. Evidence for spontaneous analogical transfer based on more than superficial similarities is sparse. Moreover, there is plenty of evidence suggesting that individuals are strongly influenced by surface similarities among source and target problems (e.g., Holyoak & Koh, 1987), and that the effect of surface similarities among problems is often to lead participants into drawing superficial analogies that do not deliver a conceptually relevant solution. Thus, there are grounds for suspecting that analogy might not be the default heuristic for tackling ill-defined problems. Recent evidence collected by Chris Bearman, a PhD student in our laboratory, suggests that encouraging individuals to make use of analogies can actively detract from their subsequent problem-solving performance. In one experiment, he reversed the order of problems used by Gick and Holyoak, giving the radiation problem as the source and a number of variants including the fortress problem as targets. When participants were given a hint to use the source solution analogically, they actually were worse at solving the target than participants who received the target alone. One point that is generally overlooked in the literature is that the radiation problem is much harder to solve than the variants (e.g., the fortress problem) that are used as analogues. It appears, then, that the encouragement to analogize led participants to use an analogical mapping strategy rather than tackling the relatively simple target problem in a more direct fashion (e.g., by trial and error or by means– ends analysis), which would have been more effective. Analogy, then, may actually be a strategy of last resort when a problem is too difficult to be tackled on its own by conventional means.


57

On-line planning and puzzle solving There are situations where individuals encounter completely novel problems for which no analogical problem and solution information may be known. The nine-dot problem (illustrated in Figure 3.1) presents such an example. Until recently, accounts of human performance have generally proposed that problem solvers impose additional and inappropriate constraints that preclude the discovery of a solution. Once these inappropriate constraints are removed, so the story goes, the problem then becomes easy to solve. The traditional Gestalt account is that the nine-dot array is unavoidably processed perceptually as a square whose boundaries should not be violated. Weisberg and Alba (1981), amongst others, point out that providing instructions to include lines that go beyond the boundary of the square does not reliably lead to solution. Instead, Weisberg and Alba propose that individuals impose an inappropriate constraint that lines should extend only between dots, a constraint derived from prior experience of dot-to-dot drawing. Curiously, they fail to notice that their critique of the Gestalt position applies equally to them. Whatever the success of these alternative accounts, the key point is that planning behaviour has not been considered as offering an account of human performance on this or other so-called “insight” problem-solving tasks. Indeed, some authors (e.g., Anderson, 2000) adopt a definition of insight problems that is based upon people’s inability to plan when solving them. For example, Metcalfe and Weibe (1987) found that measures of participants’ feeling of warmth (i.e., measures of anticipated closeness to solution) strongly predicted imminent solution of non-insight puzzles (such as the Tower of Hanoi, but not insight puzzles (such as the nine-dot problem). These results suggest that participants are unable to monitor their progress during insight problem solving, an essential prerequisite of planning behaviour. We have recently proposed an account of human performance on the ninedot problem that invokes precisely the same kinds of hill climbing strategy (i.e., move selection based upon progress made towards a hypothesized goal state) used for planning moves in non-insight problems such as the TOH (MacGregor, Ormerod, & Chronicle, 2001). Although individuals do not have a concrete and visualizable goal state against which to monitor progress, the nine-dot problem statement describes some of the goal-state properties: namely, that each dot must be cancelled and that there are four lines available to do so. These properties are enough for individuals to derive hypotheses about what might constitute “locally rational” moves; that is, moves that appear to make progress in improving the current state against the hypothesized goal properties (for related accounts of locally rational move selection, see Chater & Oaksford, 1999; Simon & Reed, 1976). So, individuals endeavour to maximize the number of dots cancelled by each move they

58

ORMEROD

make (where a move can be one or more lines). More importantly, moves are evaluated under a criterion that individuals use to judge whether the moves they currently sample are making satisfactory progress towards the hypothesized goal state – in the nine-dot problem, each move must cancel the absolute number of dots given by the ratio of remaining dots divided by remaining lines. According to our account, the standard nine-dot problem is so difficult, not because individuals impose inappropriate constraints on the lines they are prepared to sample, but because there are so many moves available that meet the criterion but that do not allow solutions to be found. Consider, for example, someone planning two lines at a time (that is, working at “two-lookahead”, a value that provided the best fit for planning behaviour across the five experiments in MacGregor et al., 2001). Figure 3.1(c) shows a typical failed attempt produced repeatedly by the majority of participants in our experiments. The criterion for each line in a satisfactory two-line move is initially 9/4, so the move must cancel at least four dots. There are 24 different two-line first moves that cancel five dots and 48 that cancel four dots. Assuming the selection of a five-dot first move, the criterion for each line in a satisfactory subsequent move is 4/2, a criterion satisfied by an available third line after all 24 moves. It is only when individuals plan their fourth and final line that they discover that they are unable to find one that meets the criterion. This is precisely the situation that seems to arise when participants generate attempts like Figure 3.1(c), and the fact that they continue to generate the same kind of attempt even when they discover failure suggests that they are driven to make this move type: the impeller here, in our view, is the late failure of the criterion for satisfactory progress, so that when individuals seek other attempts, they persevere with the same kinds of incorrect earlier move. We believe that the early experience of criterion failure is an essential prerequisite for the discovery of insightful moves. Individuals have to recognize that they are failing to make sufficient progress with the moves available within their current conceptualization of the problem space, and must seek novel non-maximizing moves. There are so many criterion-fitting moves that can be tried in the nine-dot problem (and whose failure must be remembered if they are not to be repeated), that individuals are unlikely to experience early criterion failure. There is little incentive when attempting the standard version of the problem to include move attempts that extend beyond the array of dots, since such moves would not apparently make any more progress than the moves that lie within the array. MacGregor et al. (2001, Exp. 5) presents a test of our account against theories of inappropriate constraint imposition. Figure 3.2 shows two versions of the problem, labelled (a) and (b), in which participants were given the first line of the solution, and told to cancel the remaining dots by drawing


59

Figure 3.2 The “line outside” (a) and “line within” (b) versions of the nine-dot problem tested by MacGregor, et al. (2001). Diagrams (c) and (d) show common failed attempts of participants with the “line outside” and “line within” versions, respectively (both attempts start at the top left dot).

three further lines starting from an end of the given line. All other accounts should predict greater facilitation with version (a), where the first line extends beyond the array of dots, than with version (b) where it remains within the square figure. In (a), the line violates the perceived boundary and includes a non-dot end point, thereby removing or reducing these constraints as sources of difficulty. Our account predicted the opposite result. Consider again, an individual working at two-look-ahead. The criterion for the first move in both versions is 4 (6/3 followed by 4/2), so the first move must cancel four dots. There are two available moves that meet this criterion in the “lineoutside” version, whereas there are none in the “line-within” version. Thus, we predicted that, because participants would encounter criterion failure earlier in the “line-within” version, participants would be more likely to seek alternative moves including those that extended beyond the dot array, and would therefore be more likely to solve. The results confirmed this prediction, with 45% solving the “line-outside” version and 65% solving the “line-within” version. The role of planning in insight puzzle solving is not limited to move selection. Although criterion failure is necessary for the subsequent discovery of insightful solutions, it is not sufficient, since there are, in principle, an infinite number of alternative, non-maximizing moves that participants might try once criterion failure has occurred. We suggest that, while individuals relax the requirement for moves to maximize progress once criterion failure has

60

ORMEROD

occurred, they continue to plan alternative moves by assessing the progress that they might make. Moreover, they retain a record of move attempts that fail first time round for further exploration only if the move attempt makes progress towards satisfying the criterion relative to any previous move. Also shown in Figure 3.2 are examples of failed attempts from the “line-outside” (c) and “line-within” (d) versions that occurred in our data. Both embody the conceptual “insights” to draw lines outside the dot array and to turn on nondot points. However, only the move attempt shown below the “line-within” version was a strong predictor of success on the next attempt. Participants who produced the attempt shown below the “line-outside” version were more likely to return to drawing move attempts within the dot array on their subsequent move than they were to retain the lines that they drew outside the dot array. According to our theory, they failed to capitalize on the conceptual promise of their move attempt because it made no progress relative to previous attempts, whereas the failed attempt associated with the “line-within” version cancels one more dot than any three-line attempt that remains within the dot array. In summary, it appears that individuals, when solving ill-defined puzzles, plan their move attempts by evaluating each move against a hypothesized criterion of satisfactory progress. If they fail to discover satisfactory moves, they search for emergent non-maximizing alternatives, and they are likely to retain these for further exploration if they make progress: in effect, they plan to re-use failed attempts that show promise. We have recently implemented our theory as a fully specified computational model (Ormerod, Chronicle, & MacGregor, in preparation), which implements the planning behaviours described here as a search for moves guided by a register of promising states; that is, previous failed states that have not been fully exhausted and are prioritized according to the extent to which they made progress over other attempts. As well as testing our theory with the classic nine-dot problem, we have generalized the theory to a novel insight puzzle, the eight-coin problem, which requires the discovery of moves in three dimensions (Ormerod, MacGregor, & Chronicle, 2002). The on-line local planning that we believe individuals engage in when they tackle such problems is a variant of the partial order planning approach that high ability problem solvers appear to use in tackling well-defined problems (see Davies, chapter 2, this volume).

PLANS, PLANNING AND EXPERT SKILL The problems described in the previous section are small in scale, and the consequent planning demands are relatively localized. Where the study of problem solving really becomes interesting and of relevance to explaining human activity is with large-scale realistic problems. Planning in large problem spaces can be computationally expensive. Consider the task of planning a


61

sequence of moves at the start of a game of chess, using an exhaustive planning strategy in which you evaluate the outcomes of all possible moves. There are 20 different first moves. In reply to any one of these, the opposing player would choose a move from among 20 alternatives. For each of those opposing player’s moves, you could choose a move from between 19 and 31 alternatives, depending on what the preceding moves had been. By the time you had planned and evaluated a second move exhaustively, you would have had to consider around 10,000 different board configurations. Complex and largescale, ill-defined problems defy exhaustive planning because of the combinatorial explosion of possible problem states. It seems possible, therefore, that a characteristic of domain experts that differentiates them from novices is the use of selective planning strategies. For most of the last half century, the assumption among cognitive psychologists has been that experts do not plan in real time: instead they retrieve and adapt pre-compiled plans from memory.

Is expertise just plan recall? The roots of this assumption lie in pioneering research conducted by DeGroot (1965), who challenged the folk hypothesis that the skill of grandmaster chess players is based upon an ability to plan longer sequences of moves in their heads than less skilled players. DeGroot found that there was no difference in the length of the sequence of moves that grandmasters and less skilled players planned in response to presented board positions, nor was there a difference in the time taken to select moves. Nonetheless grandmasters were able to select better quality moves. Further studies demonstrated that there was no difference between skilled and less skilled chess players in terms of general memory abilities. Thus, DeGroot concluded that chess expertise is based upon the learning and subsequent retrieval of stored knowledge about the most appropriate move to make given a specific game position. Chase and Simon (1973) confirmed this view, demonstrating that grandmasters are able to reconstruct realistic chessboards in fewer glances, placing more pieces with each glance, than less experienced players. Thus, expertise in chess appears to be based, according to these studies, on the perceptual recognition of a board position, and the retrieval of a plan that details the appropriate moves for the recognized position. Similar plan-based accounts can be found in many areas of expertise research. For example, Soloway and Erhlich (1984) proposed that expert programmers retrieve, interleave and flesh out plans in developing computer programs. Programming plans are kinds of general templates that capture generic functionality of program structures at different levels of abstraction. A recent incarnation of the programming plans approach is described by Rist (1995). Rist argues that there is no overall strategic planning or control in program design. Instead, programs are constructed through cue-based search

62

ORMEROD

and retrieval from memory (both internal and external) of plans that set up further cues to search upon until final implementation (i.e., until no further cue slots are left unfilled). As Rist states (p. 558): “The only planning mechanism is the shift of attention from one focus to another in internal and external memory.” The key cue in Rist’s view is the “focal line”, that is, an item of code or pseudo-code that directly encodes the specific goal of the plan to be retrieved from memory (e.g., a plan to accumulate a running total would be cued by the focal line “count = count + 1”). Once a focal line is retrieved, according to Rist’s account, a process of focal expansion occurs in which the remainder of the relevant code is retrieved from memory. The general notion of plans as abstract or semi-abstract templates is common to many theories of expertise. Indeed, the absence of the word “plan” from the subject index of current textbooks can be accounted for by the fact that plan-based accounts tend to be described under different jargon such as schemas (e.g., Neisser, 1976), scripts (e.g., Schank & Abelson, 1972), frames (Minsky, 1980), and Memory Organisation Packets (Schank, 1982). It is, of course, obvious that experts possess more domain knowledge than novices, and it seems likely that expert knowledge is stored in an abstract fashion that would make it amenable to plan-like adaptation as new task requirements arise. However, there is evidence to suggest that the retrieval of pre-compiled plans provides only a partial account of expert planning behaviour in ill-defined problem-solving domains. For example, Holding and Reynolds (1982) found that, when experts and novices were presented with random board positions (i.e., layouts that were not part of a realistic game and therefore could not be part of an expert’s prior knowledge of chess positions), the experts were able to make a better subsequent move. In other words, it appears that experts were using some kind of on-line planning to select the best move given the current unrecognizable (and therefore not recallable) board configuration. We have conducted research that challenges the sufficiency of a planretrieval account of programming expertise (Ormerod & Ball, 1993). We used verbal protocol analysis to examine how experts in the Prolog language (i.e., programmers with more than ten years of Prolog programming experience) constructed a solution to the relatively complex programming task (to count and calculate temporal flow statistics from a vehicle sensing device). In particular, we examined the order in which solutions were constructed. The key finding was that, while all of our nine expert programmers produced essentially the same solution in terms of its underlying structure, the order in which solutions were produced was different for every participant. Had their programming been guided by the kinds of plan structures proposed by Soloway and by Rist, then we would have expected to see focal lines emerge first, followed by template-like structures for handling the procedural control elements of the program, followed by the details. Instead, the order in which


63

programmers produced their solutions appeared to be more under the control of their current inferential goal. Each of our experts differed somewhat in terms of their backgrounds, some coming from formal specification traditions, others being jobbing programmers. This led them to emphasize different aspects of the programming task (e.g., efficiency, coding aesthetics, functionality) at different times. They all possessed the relevant programming knowledge (e.g., about the use of variables to collect intermediate results that are passed to the final output in the stopping case of the program), but the order in which this relevant knowledge was sampled was seemingly determined on-line rather than as pre-compiled execution orders.

Plans, planning and the conundrum of creative expertise Perhaps the most challenging kinds of ill-defined problem solving in which to explore planning is in creative design domains, where simple re-use of previous solutions cannot meet the necessary task condition of creativity. Design is perhaps the ultimate domain in which to explore human problem solving, a fact recognized by Simon (1981). The main issue is how problem solvers can control the process of seeking multidesign solutions in an infinite problem space, when a principle task requirement is that designers do not simply reuse previous design ideas. In the previous chapter, Davies outlined the means–ends analysis heuristic, a process by which a complex goal that cannot be achieved through a single operator application is decomposed into smaller subgoals, to a point where operators can be applied. Complex design problems require decomposition, and it is the process of managing decomposition while holding onto the earliest solution ideas that, we argue, lies at the heart of planning in expert problem solving. There are two so-called structured control strategies for decomposing complex design problems, breadth-first and depth-first decomposition, which are illustrated in Figure 3.3. In following a breadth-first decomposition approach, the problem solver decomposes the overall goal into subgoals, and then decomposes each of these subgoals in turn before trying to further decompose any one subgoal. Breadth-first decomposition has been proposed as the prescriptively optimal approach to planning in design, since it minimizes commitment to specific design solutions until the whole of a design problem has been explored (e.g., Hoare, 1978). However, breadth-first design is computationally costly, since the problem solver must maintain a mental or external register of all the design subgoals and their interrelations until the final stages of design, when all subgoals have been decomposed to a level where operators may be applied in their solution. In following a depth-first decomposition approach, the problem solver decomposes a goal into subgoals, then focuses upon these one at a time, decomposing each one as far as

64

ORMEROD

Figure 3.3 Decomposition of a hypothetical design hierarchy using three different control strategies (number order shows the order in which goals are visited under each strategy).

necessary before operator application and then returning to tackle the next subgoal. The depth-first decomposition approach underlies the means–ends analysis heuristic that appears to be applied commonly to the solution of novel problems (e.g., Newell & Simon, 1972). The advantage of depth-first design is in minimizing the number of unresolved subgoals that must be remembered as design problem solving proceeds. As well as minimizing computational load, a depth-first approach also allows for the early testing of emerging design components and enables designers to demonstrate tangible progress. A disadvantage is that early design decisions may prematurely commit the problem solver to solution approaches that do not work for later subgoals. Empirical studies of design planning have typically demonstrated that breadth-first design is associated with expert performance, while depth-first design is associated with novice performance (e.g., Adelson & Soloway, 1985; Jeffries, Turner, Polson, & Atwood, 1981). Davies (1990) has further shown that expert designers mitigate the computational load associated with breadth-first decomposition by relying much more than novices on externalization of emerging design work. This shifts the computational burden from the mental resources of the problem solver to the external task environment.


65

In a study of expert Prolog and C programmers designing solutions to a more complex version of the programming task (studied by Ormerod & Ball, 1993), we found that the experts, regardless of the language in which they were working, adopted a subtle variant of problem decomposition, which we term children-first decomposition (Ormerod, Ball, & Lang, 2003). In following this decomposition approach (which is also illustrated in Figure 3.3), designers effectively mixed breadth-first and depth-first approaches, by decomposing across all the subgoals of the main goal before focusing upon the first of these subgoals. We argue that a key component of the expertise of experienced software designers is in knowing how to manage the competing needs to maximize visible progress and early design testing while minimizing premature commitment. Planning by children-first decomposition provides a structured way for experts to achieve this balance that works well in relatively constrained design domains such as programming. The idea that designers plan the order of their problem-solving activities by applying structured control strategies has not gone uncriticized. The most radical alternative to structured decomposition is so-called opportunistic planning (e.g., Hayes-Roth & Hayes-Roth, 1979). Opportunistic accounts of design problem solving have been proposed by a number of researchers, including Guindon (1990) and Visser (1994). In both of these accounts, opportunism is characterized as a positive attribute for design, in that it provides a source for creative idea generation. It should be noted, however, that Guindon characterizes as opportunistic any design activity that does not follow a strictly breadth-first decomposition approach, while Visser makes the same contrast against a strictly depth-first approach (effectively contrasting opportunism against the implementation of a retrieved plan that is executed in depth-first fashion). We have argued that there are flaws in the analyses of both Guindon and Visser which call into question their observations of opportunistic planning (Ball & Ormerod, 1995). Instead, we argue, the design protocols they report that do not fit a single pure decomposition strategy actually reflect strategic switches between breadth-first and depth-first decomposition. Expert designers know that breadth-first design has advantages, but they also know when it is appropriate to switch to depth-first design, for example, when they want to explore a potential solution to a specific design subgoal that they believe to be particularly complex or critical to the overall design. Thus, we believe that there is an important distinction to be made between opportunism, which consists of genuinely serendipitous changes in goal focus (and abandonment of previous plans), and on-line planning, where the next moves may be determined either serendipitously or (we believe more commonly in expert design) by the pursuit of one or more global control strategies. Our studies of expert programmers (Ormerod et al., 2003) also provide evidence for local planning decisions (i.e., given a current level of focus and

66

ORMEROD

two or more goals to choose from at that level, the factors that determine the programmer’s selection). We categorized all the transitions across programming activities that did not fit a children-first decomposition strategy (18% of all transitions), according to the stated or inferred reasons for switching to a new focus. Categories of transitions accounting for more than 10% of the structure-divergent transitions were as follows: to a node whose parent has not been coded but which has been recognized – 30%; jumping back to a structured approach after a divergence – 15%; capitalizing on an analogy – 12%; debugging – 12%. Of these categories, only the “analogy” category seems opportunistic in the sense that there were serendipitous opportunities which participants capitalized upon. Software design and programming are highly skilled tasks, but one might question the extent to which they are highly creative. Indeed, the commonality among solutions produced by experts in the study by Ormerod and Ball (1993) suggests that creativity, at least in the scale of task discussed above, is limited. A recent study of expertise in the design of educational tasks enabled us to explore problem solving in a creative domain (Ormerod, Fritz, & Ridgway, 1999). Creativity is one of the primary metrics on which newly designed tasks are judged, since they must be original enough to secure copyright for the author and publisher. We set out to address two key questions in relation to planning. First, do task designers use the same kinds of problem decomposition strategy seen in other design domains or is their planning better characterized as opportunistic? Second, do task design specialists differ from non-specialists in terms of the planning strategies that they employ? We studied experts designing a novel educational task for teaching English as a second language (ESL) to meet the requirements of a brief that included information on target audience, level, duration, topic, and so forth. People who solve such problems professionally, typically the authors of ESL textbooks, may undertake months of problem-solving activity between coming up with initial ideas and finally putting their solution, a finished task, down in print. Educational task design nicely illustrates the complexities and scale of real problem-solving activities. Professional task design involves many different groups of problem solvers, including authors, publishers, teachers and target audiences. There are unlimited constraints upon task design: tasks must be short enough to fit a taught curriculum but not trivially so; they must not disenfranchise any ethnic or other group; they must not be gender specific; they must allow all learners to engage in the task activity while offering a challenge to faster learners; they must be creative and original, yet not weird or incomprehensible, and so on. Moreover, there is no single best outcome. In our study, videotapes were recorded of experienced designers and teachers (our specialist and non-specialist groups, respectively) undertaking the design of novel ESL tasks. We coded the resulting protocols according to cognitive acts (e.g., generate, evaluate, understand) and focus of activity (e.g.,


67

design brief, prior experience, analogy, emerging design, implementation). The study showed that designers and teachers differ in the tasks they produce and also in the distribution of cognitive acts across time. Each of the specialists came up with a radically different task, yet all of their solutions to the task design brief were of far higher quality than the solutions provided by teachers. Inspection of specialists’ protocols revealed four phases: idea generation, design description, expansion, and implementation. Teachers operated more on exemplar retrieval than design. Previous research suggested that designers might be less systematic in creative than in constrained domains, but precisely the opposite finding emerged. Surprisingly, there was little evidence (less than 3% of focus transitions) of opportunism by either designers or teachers (defined as a switch from one focus of activity to another in a way that is inconsistent with either breadth-first or depth-first decomposition). Instead, specialist designers worked from depth-first decomposition in the first phase to breadth-first decomposition in later phases, whereas teachers worked from breadth to depth across phases. This reverses the typical expert/ novice difference in control strategies used to manage problem solving, and provides a demonstration of strategy changes over time (see also Simon & Reed, 1976). In this domain, it makes sense for designers to begin working in a depth-first mode, since it allows them to rapidly explore and reject early ideas (a “fail fast, fail often” strategy). Once an idea has survived early testing, they switch to breadth-first decomposition, since it produces an orderly emergent design. The importance of these results for the current chapter is as follows: the only commonality that we found among expert task designers was in their approach to planning.

CONCLUSIONS In this chapter we have examined the role played by planning in solving socalled ill-defined problems. We have examined how plans can play a key role in solving ill-defined problems, in the form of expert or domain-specific knowledge of appropriate steps to execute, and as strategic search of the problem space for best value moves under a progress monitoring criterion. The effects of such planning activities can be seen both locally, in deciding what the next appropriate move might be, and globally, in deciding how to schedule the pursuit of a complex problem that decomposes into many smaller problems. Much of the work from our own laboratory has focused upon the role of strategic planning skills, in contrast to domain knowledge focused accounts of problem solving found elsewhere. Domain knowledge and strategic knowledge are both essential components of skilled problem solving, and are probably ultimately inseparable. This chapter is intended to re-emphasize the importance of strategic knowledge rather than replace domain knowledge theories.

68

ORMEROD

Given the generally accepted view that planning in advance of action – to “look before you leap” – is good practice, one might expect that planning would play a major role in current accounts of human problem solving. The study of planning has been central to research into machine learning and artificial intelligence (AI) (e.g., Nilsson, 1982). Yet, the words “plan” and “planning” do not occur in the indexes of a sample of recent textbooks on cognitive psychology (e.g., Anderson 2000; Eysenk & Keane, 2000; Groome, 1999; Reisberg, 1999; Solso, 2001). Planning has largely been overlooked in accounts of how humans tackle ill-defined problems such as insight puzzles, and plays a role that is at best secondary to memory-based accounts of expert skill in large-scale realistic problem solving. One explanation for the relative absence of planning research is as follows. Planning is a risky cognitive activity. First, we have seen how planning can be computationally expensive. Second, we have seen how plans can be misleading or inflexible. Finally, planning can be unnecessary: decisions about what move to choose at any particular point do not necessarily require the evaluation in advance of a sequence of moves. Under this view, it is perhaps surprising to argue that such a potentially costly, error-prone and wasteful cognitive activity plays an essential and positive role in human problem solving. In this chapter, we have seen how individuals are able to minimize the risks of planning and how they are able to plan in the absence of complete problem information. We have explored some of the conditions under which planning has positive outcomes in tackling ill-defined problems, as well as others where it can be detrimental to problem-solving performance. We have added another dimension, that of global versus local planning, to the distinctions between initial and concurrent planning outlined by Davies in the previous chapter, and the distinction between total and partial order planning proposed by Rattermann, Spector, Grafman, Levin, and Harward (2001). Initial versus concurrent planning refers to the position of planning activity; total versus partial order planning refers to the completeness of planning activity; and global versus local planning refers to the scope of planning activity. Plans retrieved from memory may be total order (e.g., Soloway & Erhlich’s 1984 programming plans) or partial order (e.g., Rist’s 1995 focal expansion of programming plans). Similarly, plans constructed on-line can be total order (e.g., as in the third stage of expert task design observed by Ormerod et al., 1999) or partial order (e.g., as in attempts to solve the nine-dot problem observed by MacGregor et al., 2001). In common with the way individuals appear to tackle well-defined problems, we suggest that the same kinds of goal-directed planning strategy are brought to bear with ill-defined problems. However, the view of Rattermann et al. (2001), that opportunistic planning (which they, wrongly in our view, equate with on-line planning) will be more in evidence with ill-defined problem solving, does not seem to be borne out by the evidence; though ironically, the task that


69

Rattermann et al. (2001) use to study planning behaviour is, under our account of problem definition, a well-defined problem. Indeed, there is a case for saying that, when the problem definition does not structure the activities that problem solvers can undertake, the need for structured planning increases. We suggest that individuals adopt structured planning approaches at both global and local levels. Moreover, the choice of planning strategies employed at each phase of problem solving provides a point of consistency among otherwise diverse expert behaviours in creative problem-solving domains.

REFERENCES Adelson, B., & Soloway, E. (1985). The role of domain experience in software design. IEEE Transactions on Software Engineering, SE-1. Anderson, J. R. (1993). Rules of the mind. New York: LEA. Anderson, J. R. (2000). Cognitive psychology and its implications (5th ed). New York: Worth. Ball, L. J., & Ormerod, T. C. (1995). Structured and opportunistic processing in design: A critical discussion. International Journal of Human–Computer Studies, 43, 131–151. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81. Chater, N., & Oaksford, M. (1999). Ten years of the rational analysis of cognition. Trends in Cognitive Sciences, 3, 57–65. Davies, S. P. (1990). The nature and development of programming plans. International Journal of Man–Machine Studies, 32, 461–481. DeGroot, A. D. (1965). Thought and choice in chess. Mouton: The Hague. Duncker, K. (1945). On problem solving. Psychological monographs, 58, 1–113. Eysenck, M. W., & Keane, M. T. (2000). Cognitive psychology: A student’s handbook (4th ed.). London: LEA. Gick, M. L., & Holyoak, K. J. (1980). Analogical problem solving. Cognitive Psychology, 12, 306–355. Gick, M. L., & Holyoak, K. J. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15, 1–38. Groome, D. (1999). An introduction to cognitive psychology: Processes and disorders. Hove, UK: Psychology Press. Guindon, R. (1990). Designing the design process: Exploiting opportunistic thoughts. Human–Computer Interaction, 5, 305–344. Hayes, J. R. (1978). Cognitive psychology: Thinking and creating. Homewood, IL: Dorsey Press. Hayes-Roth, B., & Hayes-Roth, F. (1979). A cognitive model of planning. Cognitive Science, 3, 275–310. Hoare, C. A. R. (1978). Communicating sequential processes. Communications of the ACM, 21, 666–677. Holding, D. H., & Reynolds, J. R. (1982). Recall or evaluation of chess positions as determinants of chess skill. Memory and Cognition, 10, 237–242. Holyoak, K. J., & Koh, K. (1987). Surface and structural similarity in analogical transfer. Memory and Cognition, 15, 332–340. Jeffries, R., Turner, A. A, Polson, P. G., & Atwood, M. E. (1981). The processes involved in designing software. In J. R. Anderson (Ed.), Cognitive skills and their acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. MacGregor, J. N., Ormerod, T. C., & Chronicle, E. P. (2001). Information-processing and insight: A process model of performance on the nine-dot and related problems. Journal of Experimental Psychology: Learning, Memory and Cognition, 27, 176–201.

70

ORMEROD

Metcalfe, J., & Weibe, D. (1987). Intuition in insight and non-insight problem-solving. Memory and Cognition, 15, 238–246. Minsky, M. (1980). K-lines: A theory of memory. Cognitive Science, 4, 117–130. Neisser, U. (1976). Cognition and reality. Freeman: San Francisco. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Nilsson, N. J. (1982). Principles of artificial intelligence. New York: Springer-Verlag. Ormerod, T. C., & Ball, L. J. (1993). Does programming knowledge or design strategy determine shifts of focus in prolog programming? In C.R. Cook, J.C. Scholtz, & J.C. Spohrer, Empirical studies of programmers 5. Palo Alto, CA: Ablex. Ormerod, T. C., Fritz, C. O., & Ridgway, J. (1999). From deep to superficial categorisation with increasing expertise. Proceedings of the 21st Conference of the Cognitive Science Society (pp. 502–506). Vancouver, August. Ormerod, T. C., MacGregor, J. N., & Chronicle, E. P. (2002). Dynamics and constraints in insight problem solving. Journal of Experimental Psychology Learning, Memory, and Cognition, 28, 791–799. Ormerod, T. C., Ball, L. J., & Lang, S. (2003). Global and local control processes of programmers. Unpublished manuscript. Lancaster University, UK. Ormerod, T. C., Chronicle, E. P., & MacGregor, J. N. (2004). A computational account of failure and success in solving knowledge-lean insight problems. Unpublished manuscript, Lancaster University, UK. Rattermann, M. J., Spector, L., Grafman, J., Levin, H., & Harward, H. (2001). Partial and totalorder planning: Evidence from normal and prefrontally damaged populations. Cognitive Science, 25, 941–975. Reisberg, D. (1999). Cognition. New York: Norton. Rist, R. S. (1995). Program structure and design. Cognitive Science, 19, 507–562. Schank, R. C. (1982). Dynamic memory. New York: Cambridge University Press. Schank, R. C., & Abelson, R. P. (1972). Conceptual dependancy: A theory of natural language understanding. Cognitive Psychology, 3, 552–631. Simon, H. A. (1981). The sciences of the artificial (2nd ed.). Cambridge, MA: MIT Press. Simon, H. A., & Reed, S. K. (1976). Modelling strategy shifts in a problem solving task. Cognitive Psychology, 8, 86–97. Soloway, E., & Erlich, K. (1984). Empirical studies of programming knowledge. IEEE Transactions on Software Engineering, SE-10, 595–609. Solso, R. L. (2001). Cognitive Psychology (6th ed.). Boston: Allyn & Bacon. Thompson, L., Gentner, D., & Lowenstein, J. (2000). Avoiding missed opportunities in managerial life: Analogical training more powerful than individual case training. Organizational Behavior and Human Decision Processes, 82, 60–75. VanLehn, K. (1989). Problem solving and cognitive skill acquisition. In M. I. Posner, (Ed.) Foundations of cognitive science. Cambridge, MA: MIT Press. Visser, W. (1994). Organisation of design activities: Opportunistic, with hierarchical episodes. Interacting with Computers, 6, 235–274. Weisberg, R. W., & Alba, J. W. (1981). An examination of the alleged role of “fixation” in the solution of several “insight” problems. Journal of Experimental Psychology: General, 110, 169–192.

CHAPTER FOUR

Working memory and planning K. J. Gilhooly School of Psychology, University of Hertfordshire, UK

INTRODUCTION The focus of this chapter will be on how planning processes are shaped by and draw on working memory in response to problem situations. This issue appears to have first been raised in Miller, Galanter, and Pribram’s (1960) classic monograph, Plans and the structure of behavior, in which they propose “a ‘working memory’ where Plans can be retained temporarily when they are being formed, or transformed, or executed” (p. 207). Subsequently, a major role for memory in planning has often been proposed (e.g., Cohen, 1996; Owen, 1997) and Cohen argued in particular that working memory is important in formulating and revising plans. Although a number of different, specific models of working memory have been developed (see Andrade, 2001; Miyake & Shah, 1999), the concept common to a range of approaches is of working memory as a limited capacity system for the maintenance and manipulation of recently acquired information, whether originating from external or internal sources. Thus, working memory would surely be expected to play an important role as an internal work space for processes of developing, maintaining and executing plans. I will focus on planning “in the head”, without the use of external memory aids, so that the demands on working memory are likely to be high. Although use of external memory aids (e.g., paper and pencil, computer-assisted planning tools) is undoubtedly common in real-life planning, little research seems to have addressed how and when external memory aids are used in planning and this may be a fruitful area for future research. 71

72

GILHOOLY

Given that planning is a response to problem situations, it is useful to consider first, what is a “problem”. The definition offered by Duncker over 50 years ago is still useful. Duncker (1945, p. 1) wrote: “A problem arises when a living organism has a goal but does not know how this goal is to be reached.” This could be rephrased as saying that there is a problem when someone has a goal but no suitable plan available from memory and so a new plan must be devised. Reitman (1965) provided a further analysis of problems as having three main components: a start state, a goal state and a set of possible actions that can lead from start state to the goal state. The problem solver’s task is to find a sequence of actions composed from the set available that will achieve the goal. The mental representation of a sequence of actions is a plan. Problems may vary in “definedness” from well defined, in which all components are clearly specified (for more on planning in well-defined domains, see Davies, chapter 2, this volume) in advance to ill-defined, where one or more components are left unspecified, e.g., “Improve the quality of life in the UK” is an ill-defined problem. Relatively little research has been carried out on planning in ill-defined tasks (although see Ormerod, chapter 3, this volume, for some examples) and I will focus on working memory and welldefined problems in this chapter. A further dimension of significance for problems is that of semantically rich (or knowledge rich) versus semantically impoverished (or knowledge-lean) problems. Semantically impoverished problems require no specialist background knowledge and are well represented by typical laboratory puzzles that all normal individuals in the culture can attempt and generally solve within a short period. By contrast, semantically rich problems require considerable background knowledge to begin tackling effectively (such as, solving chess problems; responding to emergencies in nuclear power stations). I will be discussing research on both semantically impoverished and semantically rich problems (especially chess). Contrasting these two types of tasks allows us to address the nature of expertise effects in planning which apparently allow working memory limits to be overcome. It may be useful at this point to expand on what types of processes are assumed to be involved in planning. “Planning” may be said to involve a plan production stage of mentally generating, representing, storing, evaluating and selecting possible actions sequences (plans) and a plan execution stage in which the selected plan is retrieved from long-term memory, loaded into working memory and carried out. Both stages would draw on working memory; however, most research has focused on the plan production stage and that will be reflected in the present chapter. Although planning has undoubted benefits, in enabling us to avoid potentially costly errors, it also has “mental costs”, especially in terms of loading memory resources if the planner tries to imagine and store a number of possible action sequences and contingencies in parallel or even a single extended action sequence at any one time. Indeed, in some circumstances the

4. WORKING MEMORY AND PLANNING

73

costs of planning may outweigh the benefits, as seems to be the case in Phillips, Wynn, McPherson, and Gilhooly (2001) where it was found that participants instructed to engage in planning before tackling Tower of London (TOL) problems took longer to begin moving but solved no more efficiently than participants who did little or no planning before moving. In the following sections I will briefly outline possible general strategies for developing plans, drawn from artificial intelligence (AI) research, then consider views of working memory before discussing research on working memory in planning in both a relatively simple, semantically impoverished “move” task TOL and in the more complex, semantically rich task of chess.

GENERAL STRATEGIES FOR DEVELOPING PLANS OF ACTION Research in artificial intelligence (AI) has led to specifications of very general procedures which can be applied to any well-defined problem to produce a plan (see also, Gilhooly, 1996, chapter 2, for further details). In a sense, these general procedures are plans for generating plans (or meta-plans). I will briefly set out the main outlines of such procedures as they provide possibilities for human planning processes and later discussions will draw on these concepts. There are two broad approaches which can be applied to any welldefined problem to produce plans. These are the state-action approach and the problem reduction approach.

State-action approach In the state-action approach, problems are represented as a space of states linked by actions which lead from one state to the next. A characteristic of state spaces is that they grow exponentially as one moves away from the start state. The general problem is how to find a good path to solution with minimum search. A complete search could be made by generating all possible moves and states one step ahead and then two steps ahead and so on until the goal state is found, whereupon the search process terminates and a path is identified from start to goal state. Such a breadth-first procedure is very demanding in terms of memory loading, but would eventually always produce the best path. Such a procedure is “algorithmic” in that it will always solve. A much less memory loading form of search is to take the first generated move from the start state and follow that to the next state and repeat the procedure until the goal is reached or some pre-set search limit is reached. This depth-first method is “heuristic” in that it may solve with relatively little cost in processing and storage terms, but solution cannot be guaranteed.

74

GILHOOLY

A compromise approach is known as progressive deepening. This method involves depth-first search to some pre-set limit with backing up and reapplication of depth-first search until the space has been completely searched to the pre-set depth limit or the goal has been found. If the goal is not located within the pre-set search depth limit, that limit is incremented and the process restarts. Ultimately, this method is algorithmic and, on average, will solve with less search and less memory loading than breadth first. Search of a state-space can be made more efficient and so less memory loading if there are grounds for selecting which states are most worth exploring (e.g., if they seem similar to the goal on the basis of an evaluation function which yields a suitable similarity measure). One simple heuristic method is known as hill climbing. In hill climbing one starts by generating all the immediate successor states to the start state and then one chooses that new state which gets the best score on the evaluation function for further exploration. The successors of the selected state are generated and one of them is chosen in the same way for further exploration. The process continues until, with luck, the goal state is reached. This method is heuristic rather than algorithmic and the process can lead to suboptimal solutions. A better procedure is known as best first. In this method, exploration is from the best state encountered so far, even if it was encountered some time previously. That is, a record is kept of the whole search as it evolves, and search jumps to any so far unexplored state which looks best at any particular time. This procedure is algorithmic, but does impose a potentially large memory load.

Problem reduction approach State-action approaches involve attempts to mentally manipulate problem materials from start state to goal state. The problem reduction approach is somewhat less direct and involves trying to break the overall problem down into subproblems and sub-subproblems until a level of subproblem is reached for which solutions are known. A general approach to problem reduction is known as means–ends analysis. In means–ends analysis, the solver compares the goal with the start state and identifies differences between the two states. One difference is selected and the solver searches for an action or operator in memory which serves to reduce that difference. If such an operator can be applied the solver does so and then addresses the next difference to be reduced. If the relevant operator cannot be applied because its pre-conditions are not met, then a new goal is set up of reducing the difference between the current state and the pre-conditions. The new subgoal causes a new operator to be selected and so on. The means–ends process develops a tree of goals and subgoals until subgoals are reached which can be achieved by applying a single operator.


75

A real-life example is that of making travel plans. If the problem is to travel from London to New York, the main difference is a “large distance”. The “airplane” operator is relevant to reducing “large distance” but cannot be applied immediately as various pre-conditions are not met, such as having a ticket, being at the right airport, etc. The action (or operator) “Calling travel agent” is relevant to achieving the “Ticket” pre-condition, but cannot be applied immediately if one does not have the relevant phone number. That pre-condition can be met by looking in Yellow Pages, and so on, until eventually an agent is contacted, a ticket obtained, travel to airport accomplished and the correct plane boarded. The means–ends analysis approach naturally leads to a hierarchical representation of plans. In the travel example, the top level is “Travel London– New York”. The top level breaks down into subproblems such as “Get ticket”, “Get passport”, “Travel to airport”, and each of these breaks down further into more and more specific goals. Miller et al. (1960, p. 16) strongly argued that plans were best considered as hierarchical representations “that can control the order in which a sequence of operations is to be performed”. Such hierarchical representations could also be described as involving chunks and subchunks which facilitate sequential retrieval into working memory and subsequent processing in manageable units.

WORKING MEMORY IN COMPUTATIONAL MODELS OF PROBLEM SOLVING AND PLANNING In the domain of thinking research, a background “modal” model is often assumed which derives from the classic computational work of Newell and Simon (1972). According to this approach, the cognitive system architecture consists of a vast long-term memory and a small, general purpose working memory. Processing occurs through the activation of “production rules” which lie dormant in long-term memory until triggered by the contents of working memory. (Production rules are of the form “If condition A holds, then do action B”.) The condition parts of the rules refer to working memory contents and the actions triggered can be internal (change working memory contents) and/or external (change something in the outside world). After a rule has “fired” and been applied, a new processing cycle begins in response to changed conditions in working memory, which causes further rules to be triggered, and so on. Models of processes in a range of tasks (e.g., logic, chess, cryptarithmetic) were successfully developed by Newell and Simon using this approach. It is worth noting here that essentially the same model has been applied to interpreting think-aloud performance by Ericsson and Simon (1993). According to this analysis, what people can report of their thoughts are the transient contents of working memory which constantly change as production rules fire. That is to say, people cannot report much of

76

GILHOOLY

the processes which change working memory contents, but they can report the results of such processes (the changed contents of working memory). This approach has yielded coherent results in a number of domains including planning (Gilhooly, Phillips, Wynn, Logie, & Della Sala, 1999). An interpretation of the modal model as having a sharply limited working memory capacity akin to the seven chunks of classic short-term memory (Atkinson & Shiffrin, 1968; Miller, 1956) has however been found insufficient for complex problem-solving and planning tasks by subsequent developments in artificial intelligence (AI) and simulation. Broadbent (1993) noted that artificial intelligence (AI) systems for problem solving and planning typically require manifestly larger working memory capacities than those generally believed to be available to human problem solvers. For example, Newell’s (1990, 1992) SOAR production system model can tackle a wide range of welldefined tasks but to be effective requires an unbounded working memory; whereas Anderson’s (1983) ACT* model could also apply to a range of welldefined tasks but required many more active elements in its working memory than the seven or so envisaged in the modal model. It seems from the above that more complex models of working memory, going beyond the “7+ or − 2 slots” view, will be required. Empirically based approaches to working memory offer a potential route to increased understanding, and these will be considered in the next two sections.

MULTI-COMPONENT APPROACHES TO WORKING MEMORY Baddeley and colleagues (Baddeley & Hitch, 1974; Baddeley & Logie, 1999) have developed a clear model of working memory as a multi-component system and have gathered considerable empirical support for the model, particularly through the use of dual task methodology. Originally, this system was divided into two “slave” storage systems, namely, the visuo-spatial sketchpad and the phonological loop, and a central executive which coordinates the activities of the storage systems. Recently, a third storage component labelled the episodic buffer has been added to the model (Baddeley, 2000). The phonological loop was seen as holding a limited amount of phonological or speech-based information. The visuo-spatial sketchpad was thought to hold a limited amount of visual or spatially coded information. Supporting evidence for this fractionation comes from dual task studies which show selective interference by visuo-spatial and verbal secondary tasks on memory for visuo-spatial and verbal information respectively, as well as differential patterns of impairment observed in individuals with focal brain damage (for reviews, see Baddeley & Logie, 1999; Della Sala & Logie, 1993). The central executive component is seen as having no storage functions but rather operates as an attentional controller. The episodic buffer holds


77

integrated representations of the current situation. It is assumed to be of limited capacity in terms of integrated chunks rather than number of items. Baddeley (2000) stated that the episodic buffer allows the system to “build up models representing the probable outcome of various potential actions. The episodic buffer thus plays an important role in the planning of action”. Thus, it is expected that the episodic buffer will have a major role in planning and selecting between alternative plans. However, as yet experimental manipulations of the episodic buffer have not been developed and detailed specification of its role is a target for future research. The model is subject to continuous refinement. Its components are themselves open to fractionation into subcomponents as evidence accumulates. Since the original model, the phonological loop has been fractionated into a passive phonological store and an active rehearsal process. Logie (1995) has proposed that the visuo-spatial sketchpad might better be considered more broadly as visuo-spatial working memory, with visuo-spatial tasks drawing on the central executive and two temporary memory systems, namely a passive visual cache and an active spatially based system that stores dynamic information, namely an inner scribe (see Figure 4.1). The inner scribe has been particularly linked to temporary memory for movements and movement sequences (Logie, Englekamp, Dehn, & Rudkin, 2001) and so could be expected to play a major role in planning in “move” tasks such as the Towers of London and Hanoi. The central executive is also considered open to fractionation into a number of functions (e.g., focusing attention, switching attention, activating representations in longterm memory, coordinating multiple task performance) and whether a central, general purpose controlling function is ultimately required is left as a current question for research.

Figure 4.1

Main divisions of Baddeley–Hitch–Logie working memory model.

78

GILHOOLY

SINGLE RESOURCE APPROACHES TO WORKING MEMORY Resource models provide an alternative perspective on working memory to that of structurally oriented multiple component models. The general idea of resource approaches is that working memory draws on limited pools of energy-like resources (typically “activation”) and storage and processing compete for such resources. This approach tends to use individual difference measures of working memory span which involve both storage and processing; e.g., the “reading span” measure in which participants have to read a number of sentences aloud for meaning (processing) and then recall the final words of each sentence (storage). The maximum number of sentences for which final words are recalled is the span measure. It may be noted that reading span measures do not correlate strongly with simple word span for lists of unrelated words (which may be taken to measure short-term verbal memory), but rather reflect the ability to retain earlier material in the face of subsequent unrelated processing. Just and Carpenter (1992) applied this approach in the domain of reading comprehension. They found that individual difference measures of reading span, taken to reflect working memory capacity, predicted a number of aspects of verbal comprehension, particularly when the comprehension task was demanding. Engle, Kane, and Tuholski (1999) have extended this notion to develop working memory span measures in non-verbal domains (e.g., arithmetic span) and they propose that there is a single general resource, which they label “controlled attention”, which is involved in all tasks requiring working memory (i.e., storage plus processing). Such a view suggests that performance on a wide range of problem-solving and planning tasks should be predictable from working memory span measures since problem solving and planning are assumed to place demands on working memory. Significant correlations with measures of working memory span have been reported for reading comprehension (Daneman & Carpenter, 1980, 1983); language comprehension (King & Just, 1991); reasoning (Kyllonen & Christal, 1990); and complex learning (Shute, 1991) among others. In the light of such widespread evidence of involvement of working memory spans over many domains, it has been proposed (Engle, Tuholski, Laughlin, & Conway, 1999; Kyllonen & Christal, 1990) that controlled attention may be identified with fluid intelligence. This suggests in turn that working memory span measures will be predictive over many novel tasks including planning tasks.

“MOVE” TASKS: THE TOWER OF LONDON (TOL) Small-scale, well-defined, semantically impoverished problems (puzzles) that require the movement of items from a start state to match a goal state subject to various constraints (“Move” tasks) have proven useful in the study of


79

planning and working memory. Early studies of move tasks involved the Hobbits and Orcs or Missionaries and Cannibals tasks (Simon & Reed, 1976; Thomas, 1974), the Waterjars (Atwood & Polson, 1976; Jeffries, Polson, Razran, & Attwood, 1977) and the Tower of Hanoi (TOH) tasks (Anzai & Simon, 1979). These studies suggested widespread use of heuristic methods (such as hill climbing) which would impose low loads on working memory. However, these early studies did not address the role of working memory in detail. This question has been tackled in more recent studies from our laboratory which have used a five-disc version of the TOL task. The five-disc version involves pegs of equal heights (each peg could hold all five discs) and was developed by Ward and Allport (1997). (See chapter 1 by Ward, figure 1.1; chapter 6 by Phillips, Macleod, & Kliegal, figure 6.1.) In this version, before moving any discs, the participants in our studies are normally instructed to plan the whole sequence of moves mentally before executing the planned sequence of moves. This procedure helps ensure that preplanning does indeed take place. It is plausible to suppose also that the TOL task will make demands on working memory as it requires a complex combination of processing and storage involving the generation, evaluation, selection, maintenance and execution of multistep plans. The presentation and response requirements of the TOL are visual and spatial, but it does not necessarily follow that the planning requirements load only visuo-spatial memory resources. The task could be carried out by covert verbalization of plans (Morris, Ahmed, Syed & Toone, 1993) or by visualizing a sequence of movements (Joyce & Robbins, 1991; Welsh, Cicerello, Cuneo, & Brennan, 1995). Welsh et al. (1995) cited retrospective reports by participants that they had visualized possible moves. On the other hand, Morris et al. (1993) point to brain lesion and brain activation studies (Owen, Doyon, Petrides, & Evans, 1996) which implicate left hemisphere involvement rather than right hemisphere involvement in the task. Morris et al. argue that lateralization during TOL performance is great enough to suggest verbal planning. Within the framework of the Baddeley-Hitch working memory model, Phillips, Wynn, Gilhooly, Della Sala, and Logie (1999) examined TOL performance under a range of dual tasks (articulatory suppression, spatial pattern tapping, verbal random generation, spatial random generation) in order to assess the contributions of visuo-spatial, verbal and executive components of working memory. Control, single task data were available for the TOL tasks and the secondary tasks so that possible trade-offs could be detected. The TOL tasks spanned a range of difficulty levels. It was found that articulatory suppression, which loads the phonological loop, improved performance. This apparently counter-intuitive result is consistent with other findings (Brandimonte, & Gerbino, 1993; Hitch, Brandimonte, & Walker, 1995) which indicate that interfering with verbal articulation improves performance on some visuo-spatial tasks by discouraging the use of

80

GILHOOLY

inappropriate verbal strategies and promoting the use of more suitable visuo-spatial strategies. Both verbal and visuo-spatial random generation (executive tasks) impaired performance. Simple pattern tapping (a nonexecutive task), which mainly employs the visuo-spatial slave system, had a near significant impairing effect on TOL performance. Overall, the results were interpreted as indicating involvement of visuo-spatial working memory and the central executive in effective TOL performance, with little role for the phonological loop. In a study of planning processes in the TOL task by younger and older people (Gilhooly et al., 1999), using think-aloud methods (Ericsson & Simon, 1993), we examined features of mental planning activity such as: how far ahead did participants plan; how often did errors occur in planning; when did errors tend to occur; how similar were the plans to moves actually made. A possible strategy is move selection in which all possible moves are considered one step ahead and that move selected which leads to a state most similar to the goal. However, the protocols indicated very narrowly focused search patterns with few alternative moves (usually only one) considered at each choice point in planning ahead. On the basis of the think-aloud records it was concluded that both the older and younger participants followed the same problem-reduction approach which we labelled the goal-selection strategy. Both groups tended to identify subgoals which could be attained quickly (i.e., discs which could be placed in their target positions with few moves of obstructing discs to the holding peg) and tackled those first and then tackled the next easiest subgoal, and so on. This strategy leads to very narrow search patterns, as were observed. However, the groups did differ in the success with which they applied this goal-selection strategy in planning ahead. The younger participants generally planned further ahead; they had average plan ahead distances (depths) of 6.0 mental moves as against the older participants’ average plan depths of 4.2 mental moves. The younger participants’ planned moves were more similar to their actual moves than was the case with the older participants. Part of the reason for this result was that the older participants made more errors in planning ahead (i.e., older participants proposed more moves which would not have been possible in reality). Such planning errors occur when, for example, solvers forget that they have already mentally moved a disc and plan to move it again from its original position (which no longer holds true). From these results on planning processes, it does seem that older participants are indeed less able to plan ahead on the TOL task and the most plausible reason for this outcome is that older people tend to show reduced working memory capacities (Phillips & Forshaw, 1998), which in turn would make it more difficult to plan ahead mentally. It would be expected from the above planning results that performance would be poorer in older participants when it came to actually moving the


81

discs in the post-planning stage. However, the older and younger participants differed little in number of actual moves made in solving these tasks, despite the older participants’ markedly poorer planning. Although there was no significant difference between older and younger participants on any single task in numbers of moves required, subsequent analyses did find a small advantage for younger participants when average scores over 20 TOL tasks were compared (means = 6.4 versus 6.6, F = 5.64, df = 38, p < 0.05). Why was there little age effect on the quality of solutions in the moving stage but marked age effects on the quality of planning? We suggested that when moves are actually made on the screen, participants re-run the goal selection strategy (rather than retrieving and executing a stored plan), but in the moving phase there is considerable “stimulus support” and so working memory loads are reduced compared to the planning stage. Applying the goal selection strategy on-line would only require fairly short plan lengths to be generated and retained at any one time and so the differences between older and younger participants virtually disappear. However, there were still marked individual differences in accuracy of performance over both age groups and it could still be the case that individual differences in short-term visuo-spatial and verbal working memory could affect the efficiency of online planning and so explain some of the observed variations in TOL solution efficiency. A further study by Gilhooly, Wynn, Phillips, Logie, and Della Sala (2002) examined individual differences in TOL accuracy as related to differences in verbal and visuo-spatial working memory measures together with speed and ability measures in adult participants varying widely in age. The effect of age was of interest in its own right and having a range of ages additionally helped obtain a wide range of individual differences in the other measures. The main aim was to cast light on the roles of verbal and visuo-spatial working memory components in the TOL task using an individual differences approach as a check on the previous findings based on a dual-task approach. The general methodology applied was correlational in design. We applied exploratory factor analysis and anticipated that separation would be obtained between verbal and visuo-spatial working memory factors. The main point of interest was how TOL task performance loaded on the different working memory factors; the loading pattern should be informative regarding the extent to which typical strategies draw on distinct capacities. The working memory and other measures were as follows. Total moves over 20 TOL tasks were taken as a measure of performance. Participants were also assessed on measures of fluid intelligence (Raven’s matrices; Raven, 1960); verbal short-term storage (Digit span; from Wechsler, 1981); verbal working memory span (Silly Sentence span; Baddeley, Logie, Nimmo-Smith, & Brereton, 1985); visuo-spatial short-term storage (Visual Pattern span; Della Sala, Gray, Baddeley, Allmano, & Wilson, 1999 and Corsi Block span;

82

GILHOOLY

Corsi, 1972); visuo-spatial processing speed (Manikin test; from Lezak, 1995); verbal speed (Rehearsal speed – time to repeat digits 1–10 five times); visuo-spatial working memory (Corsi Distance Estimation; Phillips & Forshaw, 1998). The Corsi Distance Estimation task is a variation of the Corsi Block task and is designed to be a visual analogue of working memory sentence span tasks. The task requires estimation of distances between boxes presented on computer screen and simultaneous storage of location of the boxes. Exploratory factor analysis using an oblique rotation method revealed three factors which were interpreted as: (1) a visuo-spatial working memory factor; (2) an age–speed factor; (3) a verbal working memory factor. The visuo-spatial and verbal factors were only moderately correlated (0.31). Performance on the TOL task loaded significantly (−0.404) on the visuospatial factor but did not load on the other factors. The negative loading of TOL score on the visuo-spatial factor indicates that a greater visuo-spatial working memory capacity was associated with lower score in terms of moves to solve. Overall, the factor analysis supported a substantial degree of separation of verbal and visuo-spatial working memory components which has been previously supported by dual task results (Baddeley & Logie, 1999; Della Sala & Logie, 1993) and by individual difference methods (Shah & Miyake, 1996). Furthermore, the loading results indicated that TOL accuracy is largely mediated by visuo-spatial working memory as against verbal working memory or cognitive speed. The TOL task was not administered as a speeded or time-pressured task and in these circumstances general speed of processing does not seem to have been an important variable when compared to visuo-spatial working memory measures. These results provide independent confirmation, using an individual differences methodology, of conclusions previously obtained by means of dualtask studies (Phillips et al., 1999): that is, the TOL task loads the visuo-spatial component of working memory. Welsh, Satterlee-Cartmell, and Stine (1999) found a similar pattern of correlations between visual working memory measures and performance on a different version of TOL more closely based on the Shallice’s original three-disc task than that used in our study. However, Welsh et al. did not include measures of verbal working memory in their study. As with Lehto’s (1996) study of the related TOH task, we did not find any evidence of verbal working memory involvement in the TOL, contrary to the view that TOL draws on verbal processes because it involves left frontal lobe function (Morris, Ahmed, Syed, & Toone, 1993). TOL deficits are more common after left than right frontal lobe lesions (Glosser & Goodglass, 1990; Shallice, 1982) and brain activation during TOL in normal participants is highest in the left frontal lobe (Morris et al., 1993; Owen et al., 1996). However, an alternative explanation is that the left frontal lobes are involved in visuo-spatial memory. This is consistent with Owen et al. (1996) who found


83

high brain activation in the left frontal cortex during visuo-spatial memory tasks in normal adult participants. Owen et al. examined brain activation during TOL solving and during a visuo-spatial memory task. Both the TOL task and the visuo-spatial memory task activated the frontal lobes, but comparisons showed greater left frontal lobe activation during the visuo-spatial memory task compared to the TOL. This suggests that much of the frontal involvement of the TOL may be defined in terms of visuo-spatial memory demands. Our think-aloud study of planning in TOL (Gilhooly et al., 1999) supported the view that the predominant strategy in TOL is one of goal selection but had left the modality of processing unspecified. The individual differences in results help specify the modality in which the predominant strategy is executed. The association of TOL accuracy with visuo-spatial working memory measures indicates that the goal selection strategy uses visuo-spatial codes and is thus executed best by individuals having high visuo-spatial capacities. This is also consistent with the findings of Phillips et al.’s (1999) dual-task study which found impairment of TOL by spatial dual tasks. Further, the major loadings of the factor most associated with TOL accuracy are those of the Corsi Blocks and Corsi Distance tasks which impose high spatial and sequential memory demands. It is plausible to suppose that the Corsi Blocks and Corsi Distance tasks heavily engage the active spatial subcomponent of visuo-spatial working memory, namely, the inner scribe. Since the TOL also loads strongly on the same factor as Corsi Blocks and Corsi Distance, it is argued that the TOL also draws on the inner scribe. It is suggested then that the predominant strategy in the TOL (i.e., goal selection) is executed using a spatial code to represent planned sequences of movements of discs and that these planned sequences are held in the active spatial rehearsal mechanism (inner scribe) of visuo-spatial working memory.

STUDIES OF PLANNING AND WORKING MEMORY IN CHESS In our discussion so far, the focus has been on semantically impoverished puzzles which can be tackled without any special prior knowledge. Chess is a good example of a semantically rich problem domain in which considerable background knowledge is needed for effective performance. It is generally estimated that about ten years of dedicated study and practice are needed to achieve expert levels of chess play (Holding, 1985). Thus, there is clearly a major role for long-term memory processes in chess skill. Compared to the best computer chess programs, human experts search relatively few possible moves and move–counter-move sequences and rely much more on recognition of familiar patterns (Chase & Simon, 1973). However, experts do search ahead and engage in somewhat deeper and broader search than novices (Charness, 1981). A progressive deepening strategy has been observed which

84

GILHOOLY

enables search to advance beyond working memory limits by storing earlier results in long-term memory (Holding, 1985). Furthermore, Ericsson and Kintsch (1995) and Ericsson and Delaney (1998, 1999) have proposed that experts can circumvent the limited capacity of working memory by rapid storage of information in long-term memory which is kept highly accessible by maintaining retrieval cues in working memory. In this way experts can maintain ready access to a larger volume of information than would be possible if working memory alone was relied upon. They propose that effectively experts use a long-term working memory which is a collection of acquired mechanisms to expand the functional capacity of working memory for materials in their area of expertise. Studies of memory for briefly presented chess positions (a task at which experts excel) have indeed shown that such material is highly resistant to typical short-term memory disrupting tasks such as counting backwards (Charness, 1976; Saariluoma, 1992) which suggests rapid storage in long-term memory. Some research has specifically addressed working memory in chess. Using dual-task methods, Saariluoma (1991a, 1992) found that visuo-spatial suppression has a marked effect on move generation, mate detection and the encoding stage of memorizing chess positions. No effects were found for articulatory suppression. Thus, a major role for the visuo-spatial sketchpad is suggested in chess play, but there is no evidence for a strong role for the phonological loop. Rather similar results were found by Baddeley (1992) and Bradley, Hudson, Robbins, and Baddeley (1987) and additionally these studies also found major effects of central executive suppression on recall of chess positions and on generating moves from given positions. Thus, it may be concluded that mental exploration of possible move sequences to find a good chess plan heavily involves the visuo-spatial sketchpad and the central executive components of working memory. The phenomenon of blindfold chess has attracted research. It is remarkable that expert players can successfully play a large number of simultaneous games (ten or more, Saariluoma, 1998) while blindfolded. They are given their opponents’ moves verbally and must store the current state of all the simultaneous games in memory. This task would seem to heavily involve long-term working memory as detailed by Ericsson and Delaney (1998, 1999). In a study of blindfold chess, Saariluoma (1991b) found large expertise effects and strong interference from concurrent visuo-spatial suppression tasks, but little effect of articulatory suppression. Thus, it would appear that the visuo-spatial sketchpad component of working memory plays a key role in the special skill of blindfold chess presumably in updating the image in response to own and opponents’ moves and in exploring new possible moves and plans. Storage of information about the games not being dealt with at any one time is assumed to be in long-term working memory.


85

CONCLUDING COMMENTS From the evidence reviewed in this chapter, it may be concluded that planning in move tasks draws on executive processes and on the visuo-spatial component of working memory with particular involvement of the inner scribe subcomponent. The phonological loop shows fewer signs of being involved (but may have a role in planning for tasks which are more verbally loaded than typical move problems). Expertise in semantically rich move tasks (e.g., chess) seems to involve use of extended working memory (long-term working memory) and this enables search of possible plans to extend deeper and wider than would be possible within the short-term working memory system. With regard to methodological issues, it may be noted that current dualtask methods indicate only that certain working memory components were involved at some time during the planning task. For the future, it would be desirable to develop ways to identify when different components are loaded as the task is worked through. In addition, more precisely targetted dual tasks would be useful to enable researchers to go beyond simply finding that, for example, the central executive is involved in a planning task, and allow identification of which particular executive functions such as “task switching”, “updating” and “inhibition” (Miyake, Friedman, Emerson, Witzki, Howerter, & Wager, 2000) are involved. Theoretically, planning strategies need more detailed specification than is often given (e.g., characterized as simply “verbal” or “visuo-spatial”). Specification in terms of information-processing steps using particular modalities and memory resources would be useful so that execution of strategies could be linked to moment-to-moment loading of working memory components. Testing of such highly specified theories would be facilitated by the development of methods for assessing the temporal sequence of loading of working memory components. The development of more detailed theories and appropriate testing methods are goals for future research on planning.

REFERENCES Anderson J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press. Andrade, J. (Ed.) (2001). Working memory in perspective. Hove, UK: Psychology Press. Anzai, Y., & Simon, H. A. (1979). The theory of learning by doing. Psychological Review, 86, 124–140. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation, (Vol. 2). New York: Academic Press. Atwood, M. E., & Polson, P. G. (1976). A process model for water jug problems. Cognitive Psychology, 8, 191–216. Baddeley, A. D. (1992) Is working memory working? Quarterly Journal of Experimental Psychology, 44A, 1–33.

86

GILHOOLY

Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Science, 4, 417–423. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. Bower (Ed.), The psychology of learning and motivation 8 (pp. 47–89). New York: Academic Press. Baddeley, A. D., & Logie, R. H. (1999). Working memory: The multiple component model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and control. Cambridge: Cambridge University Press. Baddeley, A. D., Logie, R. H., Nimmo-Smith, I., & Brereton, N. (1985). Components of fluid reading. Journal of Memory and Language, 24, 119–131. Bradley, A. C., Hudson, S. R., Robbins, T. W., & Baddeley, A. D. (1987). Working memory and chess. Unpublished report. MRC Applied Psychology Unit, Cambridge. Brandimonte, M. A., & Gerbino, W. (1993). When imagery fails: Effects of verbal recoding on accessibility of visual memories. In C. Cornoldi, R. H. Logie, M. A. Brandimonte, G. Kaufmann, & D. Reisberg (Eds.), Stretching the imagination: Representations of transformations in mental imagery. New York: Oxford University Press. Broadbent, D. E. (1993). Comparison with human experiments. In D. E. Broadbent (Ed.), The simulation of human intelligence. Oxford: Blackwell. Charness, N. (1981). Search in chess: Age and skill differences. Journal of Experimental Psychology: Human Perception and Performance, 7, 467–476. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81. Cohen, G. (1996). Memory in the real world (2nd ed.). Hove, UK: Psychology Press. Corsi, A. T. (1972). Human memory and the medial temporal region of the brain. Dissertation. Montreal: McGill University. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Daneman, M., & Carpenter, P. A. (1983). Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory and Cognition, 9, 561–584. Della Sala, S., & Logie, R. (1993). When working memory does not work: The role of working memory in neuropsychology. In F. Boller & J. Grafman (Eds.), Handbook of Neuropsychology (Vol. 8). Amsterdam: Elsevier. Della Sala, S., Gray, C., Baddeley, A., Allmano, N., & Wilson, L. (1999). Pattern span: A tool for unwelding visuo-spatial memory. Neuropsychologia, 37, 1189–1199. Duncker, K. (1945). On problem-solving. Psychological Monographs, 58, 1–113. Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999). Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence and functions of the prefrontal cortex. In A. Miyake & P. Shah (Eds.), Models of working memory. Cambridge: Cambridge University Press. Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. A. (1999). Working memory, short-term memory, and general fluid intelligence: A latent variable approach. Journal of Experimental Psychology: General, 128, 309–331. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data. Cambridge, MA: MIT Press. Ericsson, K. A., & Delaney, P. F. (1998). Working memory and expert performance. In R. H. Logie & K. J. Gilhooly (Eds.), Working memory and thinking. Hove, UK: Psychology Press. Ericsson, K. A., & Delaney, P. F. (1999). Long-term working memory as an alternative to capacity models of working memory in everyday skilled performance. In A. Miyake & P. Shah (Eds.), Models of working memory. Cambridge: Cambridge University Press. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245.


87

Gilhooly, K. J. (1996). Thinking: Directed, undirected and creative (3rd Rev. ed.). London: Academic Press. Gilhooly, K. J., Phillips, L. H., Wynn, V. E., Logie, R. H., & Della Sala, S. (1999). Planning processes and age in the five-disk Tower of London task. Thinking and Reasoning, 5, 339–361. Gilhooly, K. J., Wynn, V. E., Phillips, L. H., Logie, R. H., & Della Sala, S. (2002). Visuo-spatial and verbal working memory in the five-disk Tower of London task: An individual differences approach. Thinking and Reasoning, 8, 165–178. Gilinsky, A. S., & Judd, B. B. (1994). Working memory and bias in reasoning across the life span. Psychology and Aging, 9, 356–371. Glosser, G., & Goodglass, H. (1990). Disorders in executive control functions among aphasic and other brain damaged patients. Journal of Clinical and Experimental Neuropsychology, 12, 485–501. Hitch, G. J., Brandimonte, M. A., & Walker, P. (1995). Two types of representation in visual memory: Evidence from the effects of stimulus contrast on image combination. Memory and Cognition, 23, 147–154. Holding, D. H. (1985). The psychology of chess. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Jeffries, R., Polson, P. G., Razran, L., & Attwood, M. E. (1977). A process model for missionaries–cannibals and other river crossing problems. Cognitive Psychology, 9, 412–420. Joyce, E. M., & Robbins, T. W. (1991). Frontal lobe function in Korsakoff and non-Korsakoff alcoholics: Planning and spatial working memory. Neuropsychologia, 29, 709–723. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. King, J., & Just, M. A. (1991). Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language, 30, 580–602. Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working memory capacity. Intelligence, 14, 389–433. Lehto, J. (1996). Are executive function tests dependent on working memory capacity? Quarterly Journal of Experimental Psychology, 49, 29–50. Lezak, M. D. (1995). Neuropsychological assessment (3rd ed.). New York: Oxford University Press. Logie, R. H. (1995). Visuo-spatial working memory. Hove, UK: Psychology Press. Logie, R. H., Englekamp, J., Dehn, D., & Rudkin, S. (2001). Actions, mental actions and working memory. In M. Denis, R. H. Logie, C. Cornoldi, M. de Vega & J. Englekamp (Eds.), Imagery, language and visuo-spatial thinking. Hove, UK: Psychology Press. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–87. Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. New York: Henry Holt. Miyake, A. and Shah, P. (Eds.). (1999). Models of working memory. Cambridge: Cambridge University Press. Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive functions and their contribution to complex “frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41, 49–100. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, B. K. (1993). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367–1378. Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press. Newell, A. (1992). Precis of Unified theories of cognition. Behavioral and Brain Sciences, 15, 425–492.

88

GILHOOLY

Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Owen, A. M. (1997). Cognitive planning in humans: Neuropsychological, neuroanatomical and neuropharmacological perspectives. Progress in Neurobiology, 53, 431–450. Owen, A. M., Doyon, J., Petrides, M., & Evans, A. C. (1996). Planning and spatial working memory: A positron emission tomography study in humans. European Journal of Neuroscience, 8, 353–364. Phillips, L. H., & Forshaw, M. J. (1998). The role of working memory in age differences in reasoning. In R. H. Logie & K. J. Gilhooly (Eds.), Working memory and thinking. Hove, UK: Psychology Press. Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S. E., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology, 54A, 579–598. Raven, J. C. (1960). Guide to the standard progressive matrices. London: H.K. Lewis. Reitman, W. R. (1965). Cognition and thought. New York: Wiley. Saariluoma, P. (1991a). Visuo-spatial interference and apperception in chess. In M. Denis & R. H. Logie (Eds.), Mental images in human cognition. Amsterdam: Elsevier. Saariluoma, P. (1991b). Aspects of skilled memory in blindfold chess. Acta Psychologica, 77, 65–89. Saariluoma, P. (1992). Visuo-spatial and articulatory interference in chess players’ information intake. Applied Cognitive Psychology, 6, 77–89. Saariluoma, P. (1998). Adversary problem solving and working memory. In R. H. Logie & K. J. Gilhooly (Eds.), Working memory and thinking (pp. 115–138). Hove, UK: Psychology Press. Shallice, T. (1982). Specific impairments of planning. Philosophical Transactions of the Royal Society of London, B298, 199–209. Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial thinking and language processing: An individual differences approach. Journal of Experimental Psychology: General, 125, 4–27. Simon, H. A., & Reed, S. K. (1976). Modelling strategy shifts on a problem-solving task. Cognitive Psychology, 8, 86–97. Shute, V. J. (1991). Who is likely to acquire programming skills? Journal of Educational Computing Research, 7, 1–24. Thomas, J. C., Jr. (1974). An analysis of behavior in the hobbits-orcs problem. Cognitive Psychology, 6, 257–269. Ward, G., & Allport, D. A. (1997). Planning and problem-solving using the 5-disk Tower of London task. Quarterly Journal of Experimental Psychology, 50, 49–78. Wechsler, D. (1981). Manual for the Wechsler adult intelligence scale (Rev.). New York: Psychological Corporation. Welsh, M. C., Cicerello, A., Cuneo, R., & Brennan, M. (1995). Error and temporal patterns in Tower of Hanoi performance: Cognitive mechanisms and individual differences. Journal of General Psychology, 122, 69–81. Welsh, M. C., Satterlee-Cartmell, T., & Stine, M. (1999). Towers of Hanoi and London: Contribution of working memory and inhibition to performance. Brain and Cognition, 41, 231–242.

CHAPTER FIVE

Planning and the executive control of thought and action Geoff Ward Department of Psychology, University of Essex, UK

INTRODUCTION Planning research offers the opportunity to study how the cognitive system is able to exert “executive” or attentional control over our thoughts and actions, such that we are often able to do what we want, when we want – even when the way forward for reaching our goals is at first unclear. Ward and Morris (chapter 1, this volume) introduced a range of different uses for the terms “plans” and “planning”. It may be the case that, to some readers still, the terms plans and planning should be exclusively reserved for discussing “higher order” cognitive processes, such as novel and complex problem solving, where new solutions are needed to achieve previously unsatisfied goals (e.g., Newell & Simon, 1972). However, other readers may accept that the terms “plans” and “planning” may also encompass the cognitive control of “lower level”, pre-learned action sequences. Planning in this sense involves the selection and retrieval of established sequences of thoughts and actions that are (hopefully) appropriate to the given goals of the task in hand and the current environmental context, together with the monitoring, evaluation and troubleshooting of these plans when things go wrong. In this chapter, I will assume that the terms “plans” and “planning” may properly be applied to the discussion of both “higher order” novel and complex problem solving and to the cognitive control of “lower level” action plans. This assumption perhaps reflects the fact that my initial interest in planning research (Ward, 1993) was influenced as much by the (then emerging) literature 89

90

WARD

concerned with executive control processes (Allport, 1989, 1993; Norman & Shallice, 1986) as by the well-established literature on problem solving (e.g., Newell & Simon, 1972; Newell, Shaw, & Simon, 1958; Simon, 1978). It is perhaps worth pausing at this point to consider what is meant by executive control. Executive control processes have been thought to be necessary to control the flow of information through the cognitive system, such that when we perceive stimuli in the environment we perform the goalappropriate tasks and hence the goal-appropriate responses. By way of illustrating the rationale for the existence of executive control processes, it is worth reminding ourselves that there are many different ways that we can interact with any given stimulus in our environment. For example, as I look up from my computer, I see that I am in my dining room, and that outside the French windows it is a wet and cloudy October day. I see a glass of water on the table to my right, drink from it, and upon returning the glass, I hesitate, and then place it on a coaster that I have just noticed on the other side of the computer. In this mundane sequence of everyday actions, I could have interacted with the stimuli that I perceived in many alternative ways. I could have been tempted to open the French windows and walk outside, get up and walk to the kitchen to refill my glass, or ignore the coaster and replace the glass on the table to my right (I am right-handed, after all, and it is more convenient for me to have the glass on my right-hand side). However, I did not perform any of these alternatives, principally because I have a more important goal of writing this chapter, from which I do not wish to be distracted. The theoretical questions that this passage elicits are: How are our actions controlled? How and why do we (plan to) do one thing rather than another? How can we selectively perform one task or response, and prevent ourselves from performing others? Perhaps the most influential theoretical account of executive control is that proposed by Norman and Shallice (1986), which was adopted by Baddeley (1986) as a candidate account of the Central Executive. In this theory there are three levels of cognitive control. At the lowest level of control, perceptual stimuli trigger associated schema resulting in automatic firing of thoughts and actions. Contention scheduling is posited to resolve conflict when otherwise multiple competing responses would occur. This automatic process prioritizes otherwise competing activated schema, based on the current environmental cues and goals and the recency and frequency of past experiences. Finally, the supervisory attentional system (SAS) biases the activation of schema, favouring one over its rivals when this is necessary. Critically for planning research, the SAS was argued be used when conscious control of action was required, such as when planning or making decisions, when troubleshooting, when performing novel or poorly learned sequences of actions, when performing dangerous or technically difficult tasks, and when avoiding habitual responses.

5. PLANNING AND EXECUTIVE CONTROL OF THOUGHT AND ACTION

91

A more recent, alternative account of executive control, proposed by Monsell (1996), considers the task to be the behavioural unit of analysis. According to Monsell, behaviour is directed by a hierarchy of goals, and typically one task out of a wide range of possible tasks is currently appropriate. Within Monsell’s framework, the term, task-set, refers to a particular mode of readiness to perform a particular task. A task-set may be retrieved from memory (a task-set schema), or new task-sets may come from instruction, or by retuning or “sculpting” (Monsell & Driver, 2000) retrieved sets from memory. For relatively simple, single-step, stimulus-response tasks, the configuration of a task-set requires the mental preparation to perform that task rather than any other task. This may involve enabling and disabling connections between processing modules, setting their operating parameters, properly orienting to the stimulus inputs, and readying the effectors. Task-set configuration of multistep tasks is more complex, and requires the additional organization, sequencing and configuring of the component single steps. Most multistep tasks have options concerning both the contents and the sequencing of the components that are linked to the environmental conditions. This may involve holding later steps until conditions of earlier steps are fulfilled, in which case we must detect the trigger conditions, and then execute the task. Monitoring is also important to ensure that the intended goals and subgoals are adequately fulfilled, and troubleshooting is required if the goals are not being achieved. Possible troubleshooting actions include substituting alternative steps or setting up new embedded tasks when the goal is not being reached, or simply giving up if no solution is found. This framework has been successfully used to describe the control of “lower level” action plans (e.g., Monsell, 1996; Monsell & Driver, 2000; Rogers & Monsell, 1995), but as yet has not been directly applied to “higher order” novel and complex planning. It is interesting to compare the similarities and differences in the influential accounts (reviewed by Ward & Morris, chapter 1, this volume) of planning and executive control of action proposed by Miller, Galanter, and Pribram (1960), Newell and Simon (1972), Norman and Shallice (1986), with that of Monsell (1996). Across these accounts, there appears to be general agreement that human behaviour: (1) is an interaction between the participant and the environment; (2) is goal directed; (3) can best be investigated in terms of component units of analysis; (4) information flow within these simpler behavioural units must be controlled if coherent goal-directed behaviour is to be achieved. The accounts differ as to the preferred unit of analysis (TOTE, production rule, condition-action schema or task, respectively), and whether “higher order” control is performed by essentially the same processes and mechanisms as those that are used in “lower level” control, or whether there are qualitatively different mechanisms needed for “higher order” plans, and planning.

92

WARD

One criticism of the Norman and Shallice position is that the SAS must be better specified or even deconstructed, if the account is to be useful. There is a perceived danger in assigning a wide range of powerful, “higher order” processing to a unitary and underspecified central controller if an understanding of these processes is to be achieved (for related critiques of attention and control, see, e.g., Allport, 1980, 1989, 1993; Newell, 1980). That is, there is a danger of introducing a homunculus into the account; an intelligent “little man inside our head” that performs all the difficult decision making and planning, such that a task analysis of, say, planning would require a recursive task analysis of the homunculus (who might himself have an SAS of his own to deconstruct). Certainly Shallice (1988, 2002; Shallice & Burgess, 1993) has been amenable to the idea of fractionating the SAS, and this process is being widely pursued by experimental psychologists, neuropsychologists, and neuroscientists. Indeed, in a recent collection of articles on executive control processes, Monsell and Driver (2000) have gone so far as to banish the control homunculus, or at least attempt to replace the homunculus with multiple, specialized control functions, which are individually simpler and more tractable to research. One approach taken by Hommel, Ridderinkhof, and Theeuwes (2002) is to argue that cognitive control functions need not necessarily be considered as basic mental functions underpinned by dedicated executive systems, but may emerge as the result of the modification of more basic subroutines that take into account both internal and external factors and the appropriate selection of perceptual and response parameters. In the rest of the chapter, I shall review evidence for the existence of “lower level” action plans, consider evidence from studies both “lower order” and “higher order” planning, and attempt to draw out similarities and differences in findings arising from this research.

DO WE POSSESS “LOWER LEVEL” ACTION PLANS? Two related lines of evidence have often been cited as evidence for the existence of pre-learned “lower level” action plans (for more detailed summaries, see e.g., Baddeley, 1986; Jeannerod, 1997; Monsell, 1996; Shallice, 1988). First, it appears possible to explain the relatively common kinds of action errors that we make in our everyday life in terms of competition between and within currently appropriate (desired) and currently inappropriate (undesired, but somehow related) action plans (Norman, 1981; Reason, 1979, 1984). For example, the desire to turn right at the end of one’s road to pick up a prescription may be captured by our familiar routine (although today inappropriate) of turning left to go to work. In so doing, we perform a slip of action, and forget to visit the pharmacy. Alternatively, one may make two cups of tea (out of habit) when only making a drink for oneself, or add milk to the drink of a guest who takes their coffee black, if one normally makes tea


93

or coffee for people who take it white. From these examples, there is some suggestion that errors occur more frequently for the most overlearned procedures (where habitual but now goal irrelevant responses may be most dominant and task monitoring may be weakest). These examples also illustrate the point that when we perform actions that we do not intend, we rarely behave “randomly”. Rather we behave (albeit inappropriately) using a fluent “lower level” pre-learned action sequence that was cued by environmental stimuli. Second, deficient supervisory control coupled with intact low-level action schemas has been posited as an explanation for the more frequent and more severe patterns of slips of action observed in neuropsychological patients classified as suffering from “dysexecutive” syndrome following frontal lobe damage or disease (e.g., Shallice, 1982, 1988). Typical symptoms illustrating intact low-level schemas with inappropriate activation include behavioural rigidity, or perseveration of a prior but now inappropriate task-set, impulsiveness or utilization behaviour, and distractibility.

LOWER ORDER PLANNING: HOW DO WE CONTROL THE INITIATION OF LOWER LEVEL ACTION PLANS? The Stroop effect (Stroop, 1935) and the difficulties associated with task switching (Jersild, 1927; Spector & Biederman, 1976) are two laboratory phenomena that have been examined to investigate how well we can control the initiation and execution of low-level action plans. The classic colour word Stroop effect refers to the additional difficulty in naming the colour of the ink of a colour word relative to naming the colour of the ink of a neutral stimulus (such as a neutral word, colour patch, or string of Xs). The Stroop effect can be seen across many tasks and stimulus dimensions, and has been characterized as the additional difficulty in suppressing an unwanted task-set (Monsell, 1996) in situations where an ambiguous stimulus (such as a colour word) has a strong tendency to evoke an unwanted response or task (e.g., reading) as well as the intended response or task (e.g., naming the colour of the ink). The pattern of interference observed between the competing tasks is asymmetric. Thus, the speed of reading the word, RED, written in black ink is relatively unaffected by the fact that the word to read is a colour name. By contrast, the speed of naming the colour of the ink of the stimulus is greatly slowed if the word (as in the case of RED) is the name of an incongruent colour word. Furthermore, the degree of Stroop interference can vary from trial to trial. Cohen, Dunbar, and McClelland (1990) have successfully modelled the basic Stroop findings in a connectionist network that has competing colournaming and word-reading pathways. The connectionist weights associated with the two routes reflect the differences in training or practice in the two

94

WARD

tasks (see also MacLeod, 1991; MacLeod & Dunbar, 1988): word reading is performed more frequently than colour naming, and so the absolute values of the connection weights are larger for the word-reading pathway. This asymmetry in the strengths of the two pathways can account for the asymmetry in the pattern of Stroop effects: the product of the dominant word-reading pathway is more likely to interfere with the weaker colour-naming pathway than vice versa. In Cohen et al.’s model, specialist task demand units bias the activation of the word-reading or the colour-naming pathways, such that the word or colour stimuli dominate the activation of the output units. Furthermore, Botvinick, Braver, Barch, Carter, and Cohen (2001) have modified a related version of the Cohen et al. model (Cohen & Huston, 1994) to include conflict monitoring and resolution. Their modified model incorporates an additional conflict monitoring unit that receives input from the response layer and then feeds a function of this activation back to the task demand units. Essentially, the level of activation of the conflict monitoring unit rises when the Stroop network has difficulty in reaching a decision (i.e., when competing outputs are concurrently excited at the response layer). This increased activation is then used to moderate the activation of the task demand units, such that greater control is subsequently exerted on the network following conflict. Simulations showed that greater control was indeed exerted following conflict, such that there was less opportunity for competition between pathways, and subsequent incongruent responses were made more quickly and more accurately. This model can provide varying degrees of control on a trial-by-trial basis, and so is sensitive to the frequency of incongruent Stroop stimuli in a block (Tzelgov, Henik, & Berger, 1992) and the reported increase in latencies and accuracies that follow an error in speeded reaction time tasks (e.g., Laming, 1968). A second increasingly investigated laboratory task is that of task switching (Allport, Styles, & Hseih, 1994; Rogers & Monsell, 1995). The switch cost is the additional time necessary to perform a list of relatively simple tasks if the tasks are alternated or otherwise mixed together compared with when the tasks are blocked. It has been proposed that supervisory attentional resources are required to initiate and monitor novel sequences of tasks. The costs are greater when the alternative tasks are similar such that both can be performed on the same list of stimuli (e.g., subtracting 3 and adding 3 to lists of 2-digit numbers, Spector & Biederman, 1976), and the costs are greatly reduced if each of the stimuli on the list is compatible with only one of the tasks (e.g., subtracting 3 and providing an antonym to lists of alternating 2-digit numbers and adjectives). Switch costs have led some theorists (e.g., Monsell, 1996) to argue that some executive control processes are required to reconfigure the cognitive system in readiness for the anticipated task; a process referred to as task-set reconfiguration. Perhaps the most interesting result from this growing literature is that substantial switch costs are still found when the switch


95

between tasks is predictable and the time between the responding of one stimulus and the presentation of the next (the RSI, Response-Stimulus Interval) is greatly delayed. These residual switch costs suggest that participants cannot (Allport et al., 1994), cannot fully (Rogers & Monsell, 1995), or cannot always (De Jong, 2000) reconfigure the cognitive system in advance to prepare for the anticipated, to-be-switched task. Ward, Roberts, and Phillips (2001) have provided evidence that the mechanisms responsible for cognitive control in the Stroop tasks are rather different to the executive control needed when task switching. Ward et al. examined Stroop costs and switch costs using incongruent colour words (e.g., the word, RED, written in black ink) and incongruent numbers of digits (three 4s) as the main sets of ambiguous stimuli. Colour word Stroop costs were defined as the additional time taken to name the colour of the ink of a colour word compared with the time taken to name the colour of the ink of a string of coloured Xs (e.g., XXXX). Digit Stroop costs were defined as the additional time taken to state the number of constituent elements in an ensemble of digits (e.g., 444) compared with time taken to state the number of constituent elements in an ensemble of Xs (e.g., XXX). In addition, switch costs were calculated by subtracting the average time taken to perform individual tasks alone from the time taken to alternate between these two tasks. In three experiments, Ward et al. found very small and mainly non-significant correlations between the Stroop costs with the words and the Stroop costs with the digits, and found only small correlations between the Stroop costs and the switch costs, which sometimes failed to reach significance. By contrast, all the switch costs were consistently highly intercorrelated. These results were interpreted as evidence against unitary, general purpose executive control processes responsible for all planning and the control of action. Although Ward et al. suggested the need for common subprocesses underpinning all switch costs (such as common task-set reconfiguration), they argued that there was no need for common processes underpinning the degree of Stroop interference across tasks. Rather, in line with Cohen et al. (1990), the magnitude of the Stroop cost may reflect the ratio in training on the competing component tasks and there need not be a correlation between individuals in the ratios of reading to colour naming and the ratio of digit naming to counting. Thus, research into the Stroop effect and task switching suggests that a major difficulty in controlling which lower level action plan is initiated at any given moment in time is the ability to keep focused on goal-relevant tasks or responses when the environmental stimuli give rise to alternative or competing tasks or responses (that may be currently goal irrelevant). The level of difficulty is related to the ambiguity of the stimulus to be processed, and is more difficult if the desired response is less well practised and when the stimulus strongly evokes competing or overlearned response alternatives. Correlational studies suggest that there may be multiple specialized executive

96

WARD

control mechanisms for the Stroop effect and for task switching. In the case of the Stroop effect, interference may be the product of the training schedules on the alternative competing tasks, which may be enhanced or facilitated by task demand units that may themselves be moderated by regulatory feedback from response-conflict mechanisms. In the case of task switching, cognitive control can be exerted by activating and maintaining the desired task-set, but this may or may not be entirely or consistently under internal control. It should be noted that the use of the term “lower order” planning in this section is potentially misleading, and does not necessarily indicate that the cognitive control needed to initiate and control pre-learned action sequences is somehow low level or peripheral. For example, Botvinick et al. (2001) proposed that a brain area known as the anterior cingulate cortex, situated adjacent to the corpus callosum on the medial surface of the frontal lobe, was responsible for the monitoring and regulation of the conflict in the Stroop task and a wide range of other cognitive tasks. Botvinick et al.’s conflict monitoring units may therefore be seen as an active, central control-andadjustment mechanism. Similarly, more general production system models of executive control functions such as the Executive-Process/Interactive-Control model (EPIC, Kieras, Meyer, Ballas, & Lauber, 2000; Kieras, Meyer, Mueller, & Seymour, 1999; Meyer & Kieras, 1997a, 1997b) have been proposed that model dual-task paradigms, such as the psychological refractory period (PRP) procedure, and executive control tasks, such as task switching. Note, however, that different simulations of EPIC typically use different task-specific executives, rather than domain-general executives.

“HIGHER ORDER” PLANNING I: PLANNING WHAT TO DO TO SOLVE A PROBLEM In this next section, we shall consider how we plan what to do to solve novel and complex problems for which no pre-learned action sequences provide a solution. Such planning might be assumed to require “higher order” planning mechanisms, and as such it is interesting to consider whether the data suggest research questions, explanations, and concepts that are any different to those required for cognitive control of “lower order” plans of action (that is, for lower order planning). Although a large number of studies have investigated participants’ ability to solve novel and complex problems “on-line” (e.g., Anderson, 1993; Anzai & Simon, 1979; Karat, 1982; Newell & Simon, 1972), far fewer studies have specifically investigated participants’ ability to fully plan before problem solving. One such study was performed by Ward and Allport (1997), who used the five-disc Tower of London (TOL) task. Some puzzles used in this study were relatively straightforward and simply involved moving the discs straight to their goal positions. For other puzzles (such as those illustrated in


97

Figure 5.1), it was necessary to make a number of moves which did not directly move the discs to their goal locations but were nevertheless necessary to solve the puzzle. These moves were defined as subgoal moves, and different sequences of subgoal moves could be organized into one or more chunks.

Figure 5.1 An illustration of the display for the five-disc TOL task used by Ward and Allport (1997) showing a trial in which a tower must be disassembled (a) and a trial in which a tower must be assembled (b).

98

WARD

Participants were instructed to do all their planning in advance and only when they were ready were they to press a “ready” button and execute the sequence of moves as fast as possible, using a computer mouse. Ward and Allport found that the preparation or planning time and error rate increased with the number of different sequences (or chunks) of subgoal moves that must be made to solve different puzzles. The post-planning time needed to solve the puzzle, the playing time, increased linearly with the number of moves, suggesting that planning had indeed been carried out in advance. Ward and Allport showed that one of the major difficulties in solving these puzzles was the difficulty in selecting the best move when there was more than one promising alternative. They argued that difficulties in planning could be simplified to the application of two heuristics. The first heuristic determines which discs the participant should next try to get into position. Such discs are referred to as currently active goal discs, and are defined as the lowest discs on each of the three pegs in the goal configuration that are not already in their goal positions. There may therefore be one, two, or as many as three currently active goal discs at any one time, depending on the arrangement of the discs at the different stages in solving a puzzle. The desired goal positions for these discs are referred to as goal-relevant pegs. The second heuristic, based on means–ends analysis, determines which move best achieves these currently active goals. Clearly, a move that would immediately accomplish a goal is considered “best” and one that moved a disc from its goal position is considered “worse”. However, intermediatery evaluations such as “good” or “bad” moves are possible, and may involve clearing or obstructing the movement of currently active goal discs to their goal-relevant pegs. According to the heuristic, the most preferable move should always be taken. Ward and Allport found that planning difficulty increased when there was a tie or conflict when more than one move shared the most preferable move. In these cases it was assumed that further evaluation was necessary to resolve which move was best, and this often resulted in additional time and incorrect (suboptimal) decisions. The application of these heuristics is best seen in an experiment which examined the on-line planning times during pairs of puzzles in which the start and goal states were reversed (see Figure 5.1, for an example of such a pair). These pairs of puzzles obviously require the same minimum number of moves to solve them (since one solution path is simply the reverse of the other). Nevertheless, there was a clear advantage in planning and solving puzzles in which the goal was to assemble a single tower of discs (Figure 5.1(b)) compared with planning and solving puzzles in which the goal was to disassemble a tower of discs (Figure 5.1(a)). Ward and Allport argued that there was no goal conflict in assembling a single tower, since it was obvious in such puzzles that there was always only one currently active disc – participants need only worry about trying to get into place the lowest disc that was


99

not already in its goal position. When disassembling towers to multiple smaller towers, conflict and equivocation may arise because there is more than one currently active goal and participants are often presented with a choice of moves in which one move may help satisfy the subgoal of getting one disc into its goal position at the expense of hindering the movement of a second disc, whereas a second move would help satisfy the subgoal of getting the second disc into its goal position at the expense of hindering the movement of the first disc. Ward and Allport were thus able to show that “higher order” planning in novel and complex problem solving could be performed in advance, and that the heuristics (which could be specified as condition–action rules) and the mechanism of conflict resolution were really not very different to those proposed to control and prioritize between multiple “lower level” action plans.

“HIGHER ORDER” PLANNING II: PLANNING WHEN TO DO THINGS TO SOLVE A PROBLEM The preceding section considered how we know what to do in order to solve a problem. In this section, I outline a recent approach used to examine a second class of planning problem, in which the difficulty in achieving a particular goal depends not upon what actions need to be performed, but rather on when to perform the component actions. Take, for example, the act of cooking a large meal – say, a Sunday roast or a Christmas dinner. There may be relatively little difficulty in planning what to cook, all the ingredients may even be on the kitchen table. However, planning is still required if one wishes to make sure that all the component elements of the meal are put on to cook at the appropriate times, such that they will all be perfectly cooked and (most critically) ready on the plate at the desired time. That is, the difficulty in planning how to achieve some goals is that they require the synchronization and/or ordering of a number of component actions. Although both types of planning difficulty exist in everyday life, it is the first class of planning that has been most extensively studied in the laboratory. This is despite the fact that the difficulties in planning when to do something are experienced every day in many real-world tasks. Many of us, for example, mentally prioritize and schedule outstanding “things to do” into the timetables of our day (perhaps as we are getting up in the morning, on our way to work, or as we scurry to make coffee and turn on the computer). We may know only too well what needs to be done (the piles of exam scripts, the urgent e-mail requests for revisions to manuscripts, the pile of references to write), but the difficulty of planning our day is in evaluating the priority of each action, since different tasks take different amounts of time to complete, that the tasks come with various different deadlines, that the consequences of missing some deadlines is of greater importance than others, and that

100

WARD

there are other scheduled time commitments in the day. An extreme case of planning when to do things is the moment-by-moment problem facing air traffic controllers, who must plan the time at which aircraft should be allowed to occupy airspace or land at airports in order to maintain air safety. In addition, the difficulties involved in planning when to do something may also seem all too familiar to anyone who has helped plan a large event such as a wedding or conference: although a list of what must be ready “on the day” may be known well in advance, there is still a great deal of organization and prioritizing to ensure that component features of the event that must be booked well in advance are dealt with before those component details of the conference or wedding, which can be left to the last minute. One difficulty in experimentally analysing planning when to do things is that ideal tasks for this purpose such as cooking are lengthy and laborious, and participants who are not sufficiently expert at the component processes will vary in their necessity to plan what to do as well as when to do it. One approach is to investigate these differences, nevertheless, as this would provide important data on genuine, everyday tasks. However, the procedure adopted here is to use a simplified computer procedure to examine only the effects of synchronization on planning when to do things. Although no attempt has been made to validate this task as a “cooking task”, it is possible (at the risk of straining the analogy) to interpret the results in terms of cooking a meal in which between two and six components with varying cooking times must be put on to cook, such that all components are ready at the same time. The basic methodology is very simple, and a depiction of the problem can be seen in Figure 5.2. Participants were presented with computer displays containing a goal location (labelled “home”) and between two and six computer buttons (labelled 1 to 6) set at random distances away from the goal location. Pressing a button initiated the movement of that button towards the goal location. Once activated, a button continued moving until it arrived at the goal location. The participants’ task was to plan a sequence of button presses such that the buttons arrived at the goal location at the same time (i.e., their arrival was synchronized). The time to press the first button constituted the planning time, the difference in time between the arrival of the first and the arrival of the last button constituted the synchronization time. This methodology has a number of advantages for investigating the difficulties in planning when to do something. First, there is little or no confusion concerning the content of the task. The only possible actions to be considered are the pressing of the computer buttons, which must arrive at the goal location as close together in time as possible. The task therefore is exclusively concerned with the sequencing of the component actions, i.e., when to do something. Second, latencies can be recorded for each individual button press and the arrival of each button at the goal location, such that


101

Figure 5.2 An illustration of the display for the synchronization task. Participants must click on the computer buttons to start them moving towards the home location. The goal is for all the buttons to arrive at the goal location simultaneously.

quantitative data can be generated. In addition, each trial is relatively quick such that a large quantity of data can be collected in a typical session. Third, the basic methodology may be readily manipulated to compare the effects of experimental variables on planning performance. Figures 5.3(a) and 5.3(b) show the effects of increasing the number of computer buttons that must be synchronized and the effects of increasing the number of different goal locations that these buttons must reach on the planning time and synchronization time. The data is taken from Ward (1998). It was hypothesized that these manipulations would increase the difficulty of subgoal management in the task. As the number of computer buttons and goal locations are increased, so it becomes harder to keep clear the relationship between the component computer buttons and their intended goal locations. Thus, in this experiment, 12 participants were presented with between two and six computer buttons and between one and three goal locations. The different goal locations were labelled home A, home B, and home C. Computer buttons that were labelled with an A (e.g., A1, A2, A3, etc.) moved towards home A when activated. Buttons that were labelled with a B (e.g., B1, B2, B3, etc.) moved towards home B when activated. Buttons that were labelled with a C (e.g., C1, C2, C3, etc.) moved towards home C when activated. As can be seen from Figure 5.3, the time to press the first button (the planning time, Figure 5.3(a)) and the synchronization time (the difference in time between the arrival of the first and the last

102

WARD

Figure 5.3 The upper panels show the effects of increasing the number of computer buttons and goal locations on the planning time (a) and synchronization time (b). The lower panels show the effect of visual feedback on the planning time (c) and the synchronization time (d) using the synchronization task.

button at its respective goal location, Figure 5.3(b)) increased with the number of buttons and the number of goal locations. The three curves are diverging, showing that the difficulty in planning and synchronizing increases multiplicatively as the numbers of buttons and different locations increase. Normally, when participants click on a computer button it travels to its goal location and participants can see its progress. Under these standard


103

conditions, monitoring the movement of the computer buttons on the screen could facilitate synchronization times (and reduce planning times) since the participants need not fully plan the activation of all buttons in advance, but could monitor moving buttons on-line, and activate new buttons when they were a similar distance from the home location. This role of visual feedback was investigated by comparing performance under standard conditions with performance under conditions in which the computer buttons, once activated, travelled invisibly towards the goal location. It was anticipated that if monitoring of the moving computer buttons was taking place in the standard conditions, then planning times and synchronization times would be greater in the no feedback or invisible conditions than in the standard conditions. Thus, 12 new participants were presented with two blocks of trials. In each block, they saw a single goal location and between two and six computer buttons. In one block of trials the buttons were visible when they were pressed, whereas on the other block of trials the buttons moved invisibly to the goal location. All participants first received practice trials on both types of trial. As can be seen from Figure 5.3, the planning time (Figure 5.3(c)) and the synchronization time (Figure 5.3(d)) again increased as the number of computer buttons increased. The planning times were greater for the invisible condition compared with the standard condition, especially for the higher numbers of computer buttons. However, the synchronization times were greater for the standard condition compared with the invisible condition when there were high numbers of computer buttons. This second result was surprising and potentially very interesting. The improvement in synchronization times in the invisible condition cannot be explained simply in terms of a trade-off between planning times and synchronization times. Planning times were greater for all invisible conditions compared with all visible conditions, but the advantage in synchronization times was present only when there were many computer buttons on the screen. If the improvement in synchronization times in the invisible condition was due to a trade-off between planning times and synchronization times, then one would expect a reduction in synchronization times across all computer buttons in the invisible condition. An alternative explanation for this finding is that synchronization times in the standard condition may be impaired by participants excessively monitoring the activated computer buttons. According to this interpretation, executive control processes associated with monitoring are drawn towards the moving computer buttons, distracting attention away from the buttons that must still be activated. In the invisible condition, by contrast, all activated buttons disappear from the screen once activated, and the participant is not distracted in this way. In the standard experimental conditions, the computer buttons travel relentlessly to the goal location, and once activated need no further monitoring,

104

WARD

intervention, or troubleshooting. In the final experiment reported here, these standard conditions were compared with a “temperamental” or “sticky” button condition in which the computer buttons were prone to stop on their way to the goal location. When a computer button stopped, the participant had to reactivate the button by clicking on it again. A further 12 participants were presented with one goal location and either two, four or six computer buttons. On half of the trials, one of the computer buttons was programmed to stop having been activated after moving towards the goal location for between 0.5 and 2 seconds. Exactly which computer button was sticky was distributed evenly across the computer buttons, such that when there were four computer buttons, the button that stopped moving was the first, second, third or fourth button that was activated on an equal number of occasions. In addition to the planning time and synchronization time, this experiment also examined the reactivation time (the time to repress a computer button taken from when it stopped moving). In all conditions, the participants’ task was to make the computer buttons arrive as close together in time at the home or goal location. There was an increase in all three dependent variables with the number of computer buttons. The findings are summarized in Figure 5.4. As one would expect, there was no effect of the type of buttons (sticky or non-sticky) on the planning time (Figure 5.4(a)), as the participants did not know in advance whether the buttons were sticky or standard. Sticky buttons did increase the synchronization time (Figure 5.4(b)), but by a constant amount: there was no additional increase with greater numbers of computer buttons. This contrasted with the reactivation times (Figure 5.4(c)), which increased with increasing number of buttons. There was also a tendency (with two and six buttons) for the reactivation times to be reduced when the sticky button was the last button of the set activated, and additional tendencies (with six buttons) for the reactivation times to be reduced when the sticky button was the first button activated and greatest when the button was the penultimate button activated. Although one should be cautious when making such inferences, when these results are translated to the act of cooking they suggest (reassuringly) that the difficulties in planning and coordinating a meal increase: (1) as the number of component ingredients in the meal is increased; (2) as the complexity of the way in which ingredients are assembled is increased. However, when component ingredients are cooked out of view, the component ingredients in a meal require more planning, but are more perfectly served up together. The explanation, in cooking terms, is that ingredients that are cooking in view may distract the cook: the cook may watch the cooking ingredients and inadvertently delay the cooking of another ingredient. Finally, the time taken to respond to a problem with individual components of a meal increases with the number of components that are cooking. Problems with


105

Figure 5.4. The effects of the reliability of the computer buttons once activated on the planning time (a), the synchronization time (b), and the reactivation time (c) using the synchronization task.

the first and last components put on to cook may be more quickly resolved in complex meals than middle components, but the resulting effect of a problem on synchronization time is roughly constant whatever the complexity of the meal. Interestingly, the experimental manipulations in this “higher order” planning task may again result in findings that are interpretable in terms of the control of “lower level” action plans. Increasing the number of computer buttons and home locations can be seen as increasing the competition and conflict in determining the next button that should be activated. Planning to synchronize invisible and sticky buttons can be interpreted as decreasing and increasing the chance of “capture” of to-be-initiated actions by already initiated actions. As such, a greater understanding of executive processes involved in the control of “lower level” action plans might help theorizing in this (ostensibly) “higher order” planning task.

WHEN AND WHY DO WE PLAN? We saw in an earlier section that one current research aim in the taskswitching literature addresses whether participants can do any useful preparation in advance of an anticipated switch in task. Interestingly, this same question has recently been asked about whether participants do anything worthwhile when they “plan” during “higher order” planning tasks. Evidence for worthwhile planning comes from Ward and Allport (1997), who showed that there was a significant negative correlation between individuals’ mean planning times and their overall accuracies on the TOL task: those

106

WARD

participants who spent longest planning made the fewest errors (for related data on the original three-disc version; see also Unterrainer, Rahm, Leonhart, Ruff, & Halsband, 2003). However, Phillips, Wynn, Gilhooly, Della Sala, and Logie (1999) and Phillips, Wynn, McPherson, and Gilhooly (2001) found little effect of increasing or decreasing pre-planning time on a similar version of the same task, using secondary tasks and experimental instructions that discouraged and encouraged pre-planning, respectively. Findings such as these question whether we do anything worthwhile when we plan. We may all know people who (perhaps infuriatingly) seem to plan too little (and so, we perceive, encounter problems that could have been avoided), and others who (equally frustratingly) seem to plan too much (and so, we perceive, create problems that do not ever become realized). Generally speaking, however, most people muddle through fairly well despite their attitude to planning. This begs the question: Although we may feel that we are spending useful time planning, are we simply dithering or equivocating, wondering what to do next? Would we have performed just as well if we were instructed to cut out the planning stage and told simply to “just go for it”? O’Hara and Payne (1998) have convincingly demonstrated that the degree of planning on a task may be experimentally manipulated, and that planning increases the efficiency of problem solving and learning. In their experiments, participants were asked to solve the 8-puzzle in which 8 tiles, numbered 1 to 8, and one space, must be rearranged into a specific sequence within a 3 × 3 grid. The tiles could be moved using one of two different types of interface, such that moves were relatively easy to make using one interface, but required more key presses on the other. They argued that when moves were easy to make then participants would be encouraged to adopt an opportunistic or trial-and-error strategy. However, when the costs of making moves were increased then any cost afforded by planning would be offset if planning resulted in fewer effortful moves. O’Hara and Payne found that participants who were encouraged to perform a plan-based strategy needed fewer moves, and as training increased were faster than those participants who could more easily make opportunistic moves. Planning also resulted in improved learning, as demonstrated by transfer to related (although not different) puzzles. This strategic use of planning (based on rational analysis of optimal performance) may also help explain the differences between the results of Phillips et al. (1999, 2001) and Ward and Allport (1997). The exact TOL methodologies used in the two laboratories differed in one small but perhaps important way. Unlike the procedure adopted by Phillips et al., we used time errors in our laboratory: irritating beeping noises that occurred whenever participants took more than 2.5 seconds to make an individual move in the postplanning, playing or execution phase. These irritating beeps may provide a


107

small but sufficient cost to persuade some participants on some puzzles to plan effectively. The idea of planning as a strategy choice has recently been advocated by Ellis and Siegler (1998), from a developmental perspective (see also Davies, chapter 2, this volume for a discussion on planning as a strategy choice in healthy adults). Ellis and Siegler provided ten different reasons why children may not plan, even when planning would be useful. To summarize their discussion, they argued that the costs of planning are that it may be effortful, tedious and error-prone. Planning may also cause an unsatisfactory initial or overall delay in launching an action, and may require more activated or more enjoyable alternative activities to be suppressed. In addition, a child may be overconfident in solving puzzles without plans, or a child may believe that they have little control over their environment, or assume that the environment is benign and that someone else will plan for them. Finally, a child may simply enjoy unplanned actions. Interestingly, the ten reasons provided by Ellis and Siegler are just some of the wider developmental, social, cultural, personality, and individual differences in planning identified by Friedman and Scholnick (1998).

SUMMARY This chapter has attempted to summarize research on planning and the control of action. There is plenty of evidence that our behaviour is goal oriented and that we possess many “lower level”, well-learned plans. A great deal of the current research in the executive control literature is interested in investigating how well we are able to exert control over which plan we execute, and how much our initiation of action plans must await the presentation of the appropriate stimuli in our environment. Our understanding of “higher order” novel plans is currently receiving less attention from within this framework. Perhaps this is because it is difficult to find planning tasks that really appear to elicit novel and complex planning from participants, and it is harder still to find planning tasks for which enough data can be collected from each participant to allow quantitative data analysis and a comparison of different experimental manipulations. This chapter summarizes some recent research on planning what to do and when to do it for more or less novel and complex planning tasks. It is interesting to notice that terms used in the executive control of “lower level” planning, such as evaluation, and conflict resolution, are being used more freely in “higher order” planning tasks. It is also interesting that common questions such as whether we can effectively plan (or prepare) in advance is a current research topic for both lower level and higher order levels of planning. Future research may reveal that there is no real difference between high-level and low-level planning, merely a difference in the level of the goal in the task hierarchy, and if so, there will be a still

108

WARD

greater need for researchers interested in “higher order” planning to take interest in studies of executive control.

REFERENCES Allport, D. A. (1980). Attention and performance. In G. Claxton (Ed.), Cognitive psychology: New directions. Routledge and Kegan Paul, London. Allport, D. A. (1989). Visual attention. In M. I. Posner (Ed.), Foundations in cognitive science. Cambridge, MA: MIT Press. Allport, D. A. (1993). Attention and control: Have we been asking the wrong questions? A critical review of twenty-five years. In D. E. Meyer & S. M. Kornblum (Eds.), Attention and performance XIV: Synergies in experimental psychology, artificial intelligence (AI), and cognitive neuroscience. Cambridge, MA: MIT Press. Allport, D. A., Styles, E. A., & Hseih, S. (1994). Shifting intentional set: Exploring the dynamic control of tasks. In C. Umiltá & M. Moscovitch (Eds.), Attention and performance XV: Conscious and nonconscious information processing (pp. 421–452). Cambridge, MA: MIT Press. Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Anzai, Y., & Simon, H. A. (1979). The theory of learning by doing. Psychological Review, 86, 124–180. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Botvinick, M. M., Braver, T. S., Barch, D. M., Cater, C. S., & Cohen, J. D. (2001). Conflict monitoring and cognitive control. Psychological Review, 108, 624–652. Cohen, J. D., & Huston, T. A. (1994). Progress in the use of interactive models for understanding attention and performance. In C. Umiltá & M. Moscovitch (Eds.), Attention and performance XV: Conscious and nonconscious information processing (pp. 453–456). Cambridge, MA: MIT Press. Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of automatic processes: A parallel distributed processing account of the Stroop effect. Psychological Review, 97, 332–361. De Jong, R. (2000). An intention–activation account of residual switch costs. In S. M. Monsell & J. Driver (Eds.), Attention and performance XVIII: Control of cognitive processes (pp. 357– 376). Cambridge, MA: MIT Press. Ellis, S., & Siegler, R. S. (1998). Planning as a strategy choice. In S. L. Friedman, and E. K. Scholnick, The Developmental Psychology of Planning (pp. 183–208). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Friedman, S. L., & Scholnick, E. K. (1998). The developmental psychology of planning. Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Hommel, B., Ridderinkhof, K. R., & Theeuwes, J. (2002). Cognitive control of attention and action: Issues and trends. Psychological Research, 66, 215–219. Jeannerod, M. (1997). The cognitive neuroscience of action. Oxford: Blackwell. Jersild, A. T. (1927). Mental set and shift. Archives of Psychology, Whole No. 89. Karat, J. (1982). A model of problem solving with incomplete constraint knowledge. Cognitive Psychology, 14, 538–559. Kieras, D. E., Meyer, D. E., Ballas, J. A., & Lauber, E. J. (2000). Modern computational perspectives on executive mental processes and cognitive control. Where to from here? In S. M. Monsell and J. Driver (Eds.), Attention and performance XVIII: Control of cognitive processes (pp. 681–712). Cambridge, MA: MIT Press. Kieras, D. E., Meyer, D. E., Mueller, S., & Seymour, T. (1999). Insights into working memory from the perspective of the EPIC architecture for modelling skilled perceptual–motor and cognitive human performance. In A. Miyake & P. Shah (Eds.), Models of working memory:


109

Mechanisms of active maintenance and executive control (pp. 183–223). Cambridge: Cambridge University Press. Laming, D. R. J. (1968). Information theory of choice–reaction times. Acta Psychologica, 43, 381–412. MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrative review. Psychological Bulletin, 109, 163–203. MacLeod, C. M., & Dunbar, K. (1988). Training and Stroop-like interference: Evidence for a continuum of automaticity. Journal of Experimental Psychology: Learning, Memory, & Cognition, 14, 126–135. Meyer, D. E., & Kieras, D. E. (1997a). EPIC – A computational theory of executive cognitive processes and multiple-task performance: Part 1. Basic mechanisms. Psychological Review, 104, 3–65. Meyer, D. E., & Kieras, D. E. (1997b). EPIC – A computational theory of executive cognitive processes and multiple-task performance: Part 2. Accounts of psychological refractory period phenomena. Psychological Review, 104, 749–791. Miller, G. A., Galanter, K. H., & Pribram, K. H. (1960). Plans and the structure of behavior. New York: Holt, Rinehart, & Winston. Monsell, S. M. (1996). Control of mental processes. In V. Bruce (Ed.), Unsolved mysteries of the mind. Tutorial essays in cognition. Hove, UK: Psychology Press. Monsell, S., M. & Driver, J. (2000). Banishing the control homunculus. In S. M. Monsell and J. Driver (Eds.), Attention and Performance XVIII: Control of cognitive processes (pp. 3–32). Cambridge, MA: MIT Press. Newell, A. (1980). Reasoning, problem-solving, and decision process. In R. Nickerson (Ed.), Attention and Performance VIII: Control of Cognitive Processes (pp. 693–718). Cambridge, MA: MIT Press. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Newell, A., Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem solving. Psychological Review, 65, 151–166. Norman, D. A. (1981). Categorisation of action slips. Psychological Review, 88, 1–15. Norman, D. A., & Shallice, T. (1986). Attention to action: willed and automatic control of behaviour. In R. J. Davison, G. E. Schwartz, & D. Shapiro (Eds.), Consciousness and selfregulation (Vol. 4). New York: Plenum Press. O’Hara, K. P., & Payne, S. J. (1998). The effects of operator implementation cost on planfulness of problem solving and learning. Cognitive Psychology, 35, 34–70. Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S. E., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology, 54A, 579– 598. Reason, J. T. (1979). Actions not as planned: The price of automatisation. In G. Underwood & R. Stevens (Eds.), Aspects of consciousness (Vol. 1). London: Academic Press. Reason, J. T. (1984). Absentmindedness and cognitive control. In J. E. Harris and P. E. Morris (Eds.), Everyday memory, actions and absentmindedness. London: Academic Press. Rogers, R. D., & Monsell, S. M. (1995). Costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124, 207–231. Shallice, T. (1982). Specific impairments in planning. Philosophical Transactions of the Royal Society London, B298, 199–209. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T. (2002). Fractionation of the supervisory system. In D. Stuss & R. T. Knight (Eds.), Principles of frontal lobe function (pp. 267–277). Oxford: Oxford University Press.

110

WARD

Shallice, T., & Burgess, P. W. (1993). Supervisory control of thought and action. In A. D. Baddeley and L. Weiskrantz (Eds.), Attention: Selection, awareness, and control: A tribute to Donald Broadbent. Oxford: Oxford University Press. Simon, H. A. (1978). Information-processing theory of human problem solving. In W. K. Estes (Ed.), Handbook of learning & cognitive processes: V. Human information (pp. 271–295). Oxford: Lawrence Erlbaum Associates Ltd. Spector, A., & Biederman, I. (1976). Mental set and mental shift revisited. American Journal of Psychology, 89, 669–679. Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643–662. Tzelgov, J., Henik, A., & Berger, J. (1992). Controlling Stroop effects by manipulating expectations for color words. Memory and Cognition, 20, 727–735. Unterrainer, J. M., Rahm, B., Leonhart, R., Ruff, C. C., & Halsband, U. (2003). The Tower of London: The impact of instructions, cueing, and learning on planning abilities. Cognitive Brain Research, 17, 675–683. Ward, G. (1993). An experimental investigation of executive processes. Unpublished D.Phil. thesis. University of Oxford. Ward, G. (1998, Brighton). Planning to synchronise events. Paper presented at the Annual Conference of the British Psychological Society, March. Ward, G., & Allport, D. A. (1997). Planning and problem-solving using the 5-disk Tower of London task. Quarterly Journal of Experimental Psychology, 50A, 49–78. Ward, G., Roberts, M. J., & Phillips, L. H. (2001). Task-switching cost, Stroop-costs, and executive control: A correlational study. Quarterly Journal of Experimental Psychology, 54A, 491–511.

CHAPTER SIX

Adult ageing and cognitive planning Louise H. Phillips Department of Psychology, University of Aberdeen, UK

Mairi S. MacLeod Department of Psychological Medicine, Gartnavel Royal Hospital, Glasgow, UK

Matthias Kliegel Department of Gerontopsychology, Institute for Psychology, Zurich,

Switzerland

AGE, THE FRONTAL LOBES AND EXECUTIVE FUNCTION Neuropsychological theories of adult ageing have recently emphasized the localization of age-related changes in the frontal lobes of the brain, with implications for age changes in executive functions of cognition (e.g., West, 1996). Neuroimaging studies tend to show that the volume of the frontal lobes decrease more than other cerebral areas with age (e.g., Raz, 1996), and also that blood flow to the frontal lobes is reduced with age (Gur, Gur, Obrist, Skolnick, & Reivich, 1987). Ageing may particularly influence functions dependent on the dorsolateral prefrontal cortex (Phillips, McPherson, & Della Sala, 2002). Executive control processes of cognition associated with dorsolateral prefrontal functioning such as planning, attentional switching and inhibition have often been shown to be impaired in older adults (e.g., Andrés & Van der Linden, 2000; Gilhooly, Phillips, Wynn, Logie, & Della Sala, 1999; Kray & Lindenberger, 2000). It has often been argued that one of the key functions of the frontal lobes of the brain is planning. In particular, it has been argued that disorders of planning are among the most disabling features of frontal lobe injury, because planning abilities are essential in many everyday tasks such as cooking or shopping, and are also important in many occupational settings. Martin and Ewert (1997, p. 578) argue: “Because the ability to plan and follow up on 111

112

PHILLIPS, MACLEOD, KLIEGEL

the results of a plan can be seen as the most important requirements of independent living of the elderly, planning ability is of particular interest to psychogerontologists.” In this chapter, we review the effects of adult ageing on planning functions. Two types of planning tasks have tended to predominate the ageing literature. First, the more structured and abstract Tower of London (TOL) task (see Figure 6.1) has been a popular method of assessing planning function as it affords several methods of quantifying planning ability. Second, a range of more open-ended planning and scheduling tasks have been used with the aim of assessing more realistic cognitive performance. We will consider the nature of age deficits in planning tasks, the consequences of age-related changes in

Figure 6.1 Five-disc TOL task. Letters represent different coloured discs: R = red, B = blue, Y = yellow, P = purple, G = green. Source: Reproduced from ‘Age, Working Memory, and The Tower of London Task’ by Louise Phillips, Ken J. Gilhooly, Robert H. Logie, Sergio Della Sala, Valerie E. Wynn, The European Journal of Cognitive Psychology, vol. 15(2) (2003) pp. 292. http: //www.tandf.co.uk

6. ADULT AGEING AND COGNITIVE PLANNING

113

planning for everyday functioning in older adults, and how planning is related to other cognitive changes with age.

EFFECTS OF AGEING ON THE TOWER OF LONDON (TOL) PLANNING TASK In the five-disc version of the TOL used in our studies, the task differs from the original design by Shallice (1982); not only are there more discs but the rods are identical in length. The procedure also used instructions to plan the whole sequence of moves that must be carried out mentally (the “plan phase”) before executing the planned sequence in a “move phase”. This methodology attempts to separate out the mental planning phase of the task from the subsequent motor execution of the plan, although in fact, people are unlikely to attempt a full mental plan of a long sequence of moves, and will tend to carry out plan completion, correction and modification on-line during the “move phase” (Phillips, Wynn, McPherson, & Gilhooly, 2001). The TOL is likely to make extensive demands on working memory, because efficient generation, execution and modification of a cognitive plan demand simultaneous storage and processing of numerous subgoals. There is evidence for the involvement of working memory in the TOL (Welsh, SatterleeCartmell, & Stine, 1999), particularly visuo-spatial working memory (Phillips, Wynn, Gilhooly, Della Sala, & Logie, 1999). Neuroimaging studies provide consistent evidence for the involvement of the frontal lobes, particularly dorsolateral prefrontal regions, in the TOL task (e.g., Baker et al., 1996; Owen, 1997; Owen, Doyon, Petrides, & Evans, 1996). A number of studies have been carried out investigating the effects of age on performance of the TOL task, using different versions of the TOL which vary in the number of discs present, minimum number of moves for solution, and number of subgoals to be resolved. Older adults tend to be slower than younger adults on the TOL, and need more moves to solve TOL trials, which suggests less efficient planning (Allmano, Della Sala, Laiacona, Pasetti, & Spinnler, 1987; Andrés & Van der Linden, 2000; Crawford, Bryan, Luszcz, Obonsawin, & Stewart, 2000; Gilhooly et al., 1999; MacLeod, 2001; Robbins et al., 1998). Older adults tend to spend more time carrying out the mental planning stage of the TOL task, and this is particularly apparent in the early, easier TOL trials (Gilhooly et al., 1999; Robbins et al., 1998). It is possible that this might reflect higher motivation amongst older adults to formulate more detailed plans when a relatively small number of moves are required and therefore can be carried out within limited working memory capacity. Older adults are always slower at carrying out the move phase of the TOL task, i.e., they take longer to make each move required for solution. This may partially reflect motor slowing with age, but, as suggested above, the move phase of the TOL task is also likely to include some on-line planning and so

114


slower move production with age may indicate less efficient planning. Older adults tend to need more moves to solve TOL trials, although this effect is not significant in all studies. Age differences in the number of moves required to solve TOL trials does not interact with the complexity of the TOL trials (e.g., Andrés & Van der Linden, 2000; Gilhooly et al., 1999; Robbins et al., 1998). This indicates that increasing the number of subgoals that must be resolved in a TOL trial does not cause particular problems for older adults. Patients with frontal lobe injuries tend to have particular problems on the Tower of Hanoi (TOH) task in their first encounter with goal–subgoal conflict (Morris, Miotto, & Feigenbaum, 1997), while older adults do not show particular difficulty in dealing with goal conflict (Gilhooly et al., 1999). The findings consistently suggest that older adults do not have particular problems in planning more lengthy TOL trials, and this may reflect the fact that most participants when carrying out the TOL task under standard instructions tend only to plan a few moves ahead, then execute those moves, before continuing to plan another subgoal. Mental planning in the TOL task can be further investigated by asking participants to plan a solution to the task and then specify how many moves are needed to solve the trial before making any moves. One indication of how accurately people are able to formulate and monitor a multistage mental plan can be obtained by comparing the number of moves estimated to be necessary by the participant to solve a trial with the actual minimum moves required. Phillips, Smith, and Gilhooly (2002) found that older adults were generally less accurate than young in estimating from a mental plan how many moves were required to solve a TOL trial. In particular, older participants were more optimistic in their move estimation; in other words, they tended to underestimate how many moves were needed to solve TOL trials. Furthermore, the effect of age on move estimation was not reduced by covarying planning time, which suggests that the tendency of older adults to be over-optimistic is not due to insufficient time spent planning, and may instead reflect a general pattern of optimistic metacognition judgements in older adults (see e.g., Connor, Dunlosky, & Hertzog, 1997). However, another possibility is that older adults might have been less able to make full-length plans in available working memory capacity, and this resulted in their mental plans being shorter. Gilhooly et al. (1999) investigated the nature of planning in the TOL task in more detail, through the use of verbal protocols. Participants were asked to “think aloud” during the planning and move phases of 20 TOL trials. The verbal protocols made were then transcribed to indicate the plans made, as well as any corrections or changes to the plans. From the protocols, planned sequences of mental moves were extracted. Executed move sequences were also recorded on the computer. Results indicated that the plans articulated by older adults were shorter and more error prone than the plans made by young


115

adults. Despite the large age-related deficit in accuracy of verbal plans, the subsequent execution of moves in this study showed minimal differences between old and young. In other words, the older adults were particularly impaired in the ability to prepare and verbally report a mental plan, but showed reasonably good ability to plan on-line. A measure of correspondence was obtained between the verbally outlined plans and the actual moves that were subsequently executed, and this indicated that older participants showed a particularly poor match between planned and executed moves. Gilhooly et al. found that there was no evidence of age differences in the strategies used to carry out planning on the TOL, and instead argue that the age deficits in mental planning on the TOL were attributable to declines in working memory capacity. Analysis of partial correlations suggests that some of the age variance in accuracy of solving TOL trials can be explained by differences in visuo-spatial working memory capacity (Phillips & Forshaw, 1998). Phillips, Gilhooly, Logie, Della Sala, and Wynn (2003) investigated the involvement of the various components of the Baddeley and Hitch (1974; Baddeley, 1986) working memory model in age differences in the TOL task. Articulatory suppression was used to load verbal rehearsal, pattern tapping to load spatial rehearsal, and verbal and spatial random generation tasks were used to load the “central executive” component of working memory in addition to the domain-specific rehearsal systems. Younger and older adults carried out the TOL task both on its own and in combination with these four secondary tasks. Older adults showed generally high levels of interference between the secondary tasks and concurrent TOL, while younger adults showed a more specific interference pattern between TOL and the random generation tasks. These results indicate that for younger adults the TOL task heavily depended upon executive resources, whereas for older adults the task heavily loaded on both slave systems as well as executive resources. These results also suggest that for younger adults, planning performance on the TOL is more specifically dependent upon executive functioning, whereas for older adults the TOL task loads many cognitive systems.

FORMULATING COMPLEX PLANS IN THE LABORATORY A number of studies have looked at the effects of ageing on tasks of paper and pencil plan formulation. In these tasks, participants are required to achieve a complex overall goal (e.g., planning a holiday) through prioritizing and ordering a range of subtasks. These paradigms assess the ability to formulate a plan without subsequent execution of the plan. Hayes-Roth and Hayes-Roth (1979) devised a paper and pencil errand planning task, in which people must prioritize and schedule a set of tasks, using a map. Bisiacchi, Sgaramella, and Farinello (1998) used a variant of this task to examine the

116


planning performance of a group of younger (20–30), older (70–80) and very old (80+) participants. They asked participants to plan the execution of a sequence of ten errands within a fictitious environment. Various constraints were outlined in the instructions, such as the opening hours of shops, and the time taken to reach certain locations on the map. The oldest group made more intrusion errors than the younger participants (i.e., they carried out a task that was not specified in the instructions). The authors attribute these failures to a problem with controlling the contention scheduling mechanism in the supervisory attentional system (SAS) (Norman & Shallice, 1986). MacLeod (2001) reports the effects of age on a range of planning tasks, including a “party planning task”. This task is illustrated in Box 6.1, and was developed from one described by Pentland, Todd, and Anderson (1998), who assessed the impact of head injury on planning ability in children. In the adult version of the party planning task (MacLeod, 2001), participants were asked to allocate 19 subtasks to three people who were organizing the party, including putting each subtask in a particular timeslot. Errors of planning were scored (following Shallice & Burgess, 1991) in terms of rule breaks, task failures, interpretation errors, and inefficient planning. A total of 95 adults ranging in age from 16 to 79 completed the party planning task. Results indicated age differences in a number of performance measures. Older adults were more likely to break the task rules, and in particular were less likely to allocate sufficient time to carry out a task within the plan. In addition, older adults were particularly poor at allocating appropriate travelling times (walking and driving) within their plan. Omission of a subtask from a plan (classified as task failures) was more common in older compared to young adults. Older participants were much more likely to omit one of the tasks from the final plan compared to younger participants, and the omitted tasks tended to be those which may have been perceived to be peripheral to the central party planning task (for example, failing to allow the three party organizers to have lunch or wash). This suggests that older adults mainly focused on the salient aspects of the plan and failed to take the finer details into their plan formulation. There was also a trend towards older adults making more interpretation errors, and most commonly this involved allocating a task to the wrong person. Finally, older adults showed more planning inefficiencies than young, although this age difference did not reach significance. The most common inefficiency was not collecting the car on the way into town, which would have resulted in an extra journey. Another complex planning task that was developed to look at patient planning performance was the Virtual Planning Task (Miotto & Morris, 1998). The task was presented as a boardgame, where participants must plan for a trip abroad by choosing relevant cards and placing them on slots representing days of the week. A set of distracter activities are also included, which relate to the current context (such as “Go to a picnic with your


117

friends”), rather than the relevant items relating to the trip abroad (such as “Call the travel agency to book the air flight tickets”). Frontal patients scored more poorly on the task than the controls. In particular, they tended to include in their plan irrelevant activities related to the current situation. MacLeod (2001) investigated the effect of age on performing this holiday Box 6.1

Cocktail party planning task used by MacLeod (2001) Instructions

It is 11.30 am on Saturday morning and Mel has invited some friends to come round at 7 pm that evening for a cocktail party and some light desserts. Mel’s two friends, John and James, have just arrived at Mel’s house on the bus (they cannot drive). They are both able to help Mel complete the list of “Things to do” before the party starts. However, none of the tasks are in order so you must delegate each of the tasks to Mel, John and James. Write down your plans on the sheets available using as many sheets as you wish to help you, by writing the one word abbreviation for each task in the appropriate column. Please remember to mark “Masterplan” in red pen on the final plan. You should imagine that Mel, John and James are all at Mel’s house and that they receive your plan at 11.30 am. A map of the area is also provided to help you. The walking/driving distances are also displayed on the map. List of things to do before party starts John is due to play tennis between 2 and 3 o’clock at the sports centre. [TENNIS] Mel’s car has to be picked up from the garage at 1.30 (repairs paid for). [CAR] Five chairs need to be borrowed from the neighbours who will be in between 5 and 5.30 pm (15 minutes). [CHAIRS] The living room should be hoovered after it has been tidied (20 minutes). [HOOVER] The cocktails need to be mixed (1 hour 30 minutes). [COCKTAILS] James forgot to bring the stereo around that he promised to bring. Mel will have to drive James home to collect it. [STEREO] Mel needs 1 hour to get washed and dressed before the party starts. [WASH] Mel has got no money and John and James cannot lend her any. She can withdraw money at the nearest bank machine at the supermarket in town. [MONEY] A cake needs to be bought from the bakery and ordered at least 2 hours in advance before it can be picked up and paid for. The bakery shuts at 2.30 pm. [PHONE BAKERS] [COLLECT CAKE] James has to go to a football match from 3 to 4.30 at the sports centre. [FOOTBALL]

118


Mel forgot to ask if any of the guests were vegetarians, so she needs to ring them all to check beforehand so she knows what food to buy (30 minutes). [RING FRIENDS] The glasses and plates need to be put out after the living room has been hoovered (15 minutes). [CROCKERY]. Somebody needs to whip the cream and lay out the crackers and cheese (1 hour). [PREPARE FOOD] Ice cubes need to be made (15 minutes to prepare, and 7 hours to set). [ICE CUBES] There is no food or alcohol in the house. The party food needs to be bought at the supermarket in town (1 hour needs to be allocated for time spent in the supermarket). [SUPERMARKET] A pizza needs to be ordered at least 1 hour in advance for Mel, John and James to eat for lunch. They will take half an hour for their lunch break. James will pay for the pizza. [ORDER PIZZA] [LUNCH] The living room needs to be tidied (30 minutes). [TIDY ROOM]

planning task in 96 participants. Older adults tended to make more errors on the planning task. They were more likely to include distracter tasks in their holiday plan, and their selection of irrelevant tasks suggested a potential problem with inhibitory control related to old age (Hasher & Zacks, 1988; Nielson, Langenecker, & Garavan, 2002). In the Virtual Planning Task, participants were asked to make a plan, check their plan, and then make any amendments to improve it. Age differences in plan accuracy remained significant after the “amendment” phase of the task. Indeed, the correlation between age and errors made on the Virtual Planning Task was actually higher following the plan correction phase (correlation between age and initial errors was 0.306, and between age and errors following the amendment phase was 0.480). This suggests that, with increasing age, people tend to become less able to take advantage of the opportunity to monitor for errors in a plan. Correlational analyses revealed relatively weak relationships between performance on the Virtual Planning Task and measures of attention (from the Test of Everyday Attention; Robertson, Ward, Ridgeway, & Nimmo-Smith, 1994). Neither attentional measures nor cognitive speed explained a substantial proportion of the age variance in performance on the Virtual Planning Task. In a study of 48 older adults aged from 65 to 97, Martin and Ewert (1997) investigated the role of working memory capacity and inhibition of irrelevant information in solving a complex planning task (planning a holiday trip). The planning task was split into an information gathering phase and a planning phase. They argued that in more naturalistic planning tasks, usually more information is provided than is needed, and that the information varies in relevance. In line with the view that adult ageing results in an inhibitory


119

deficit, Martin and Ewert argued that less efficient inhibition might result in an increased amount of task-irrelevant information overloading the working memory system and interfering with relevant information. In addition, some information which is relevant in the information-gathering phase – e.g., connected to a rejected plan version – might become irrelevant in the finished plan. Correlational analyses indicated that participants who remembered proportionally more relevant than irrelevant information in the task tended to show better planning. Martin and Ewert argued that this reflects good ability to inhibit irrelevant information which improves planning performance. However, poor performance on the Stroop task of inhibiting colour word reading did not relate to planning performance, suggesting that the relationship between inhibitory deficits and planning might be domain specific. Planning performance was predicted by working memory capacity as measured by the backward digit span task. It was concluded that older adults with poor working memory and/or inhibitory deficits were likely to have particular problems with complex planning tasks.

AGE AND ACTION PLANNING: SIX ELEMENTS TASK (SET) The planning tasks outlined in the previous section all tap the ability to formulate a plan, but do not require participants to go on to execute that plan. One task that looks at the ability to schedule and monitor the execution of an action plan is the Six Elements Task (SET). Box 6.2 illustrates a version of this task. Shallice and Burgess (1991) devised the SET because they wanted to develop tasks that were relatively open-ended, where participants must decide for themselves when to initiate and terminate subtasks. The SET requires that six individual subtasks are scheduled and performed within a time limit according to the constraints of task rules. An important aspect of the SET is that the instructions emphasize that the beginning of subtasks contain the most important items. This means that good performance on the task would comprise attempting a small number of items at the beginning of each subtask rather than completing any of the subtasks from beginning to end. Shallice and Burgess (1991) found that patients with frontal lobe damage were likely to show errors on the SET such as inaccurate plan formulation, faulty marker creation, or poor goal articulation. Garden, Phillips, and McPherson (2001) reported performance on the SET in younger (aged 31 to 46) and middle-aged (aged 53 to 64) adults, and the instructions used for this task can be seen in Box 6.2. The age groups did not differ in the efficiency of scheduling the subtasks. However, younger adults were more likely than middle-aged adults to break the task rule that the beginning items of a subtask were more important than those towards the end. There was therefore no evidence in this sample of an age-related decline in ability to plan multiple subgoals. The middle-aged group was found to show deficits on

120


Box 6.2

The version of the Six Elements Task (SET) of planning used by Garden, Phillips, & McPherson (2001)

We would like you to carry out three different tasks. Each task has two parts (A) and (B): Task 1: (A) Dictate into the tape recorder a brief account of your journey TO here. (B) Dictate into the tape recorder a brief account of your intended journey FROM here. Task 2: (A) You are to write down the names of as many pictures as you can (in order) in set A. (B) You are to write down the names of as many pictures as you can (in order) in set B. Task 3: (A) You are to solve arithmetic problems in set I. (B) You are to solve arithmetic problems in set II. You have a total of 15 minutes to complete the test. Task rules 1 You are not allowed to carry out (A) and (B) of the same task directly after each other. For example: If you do Task 1 (A), you may not do Task 1 (B) directly after. Instead you would do Task 1 (A), then attempt Task 2 (A) or Task 2 (B) or Task 3 (A) or Task 3 (B). 2 Task 1, Task 2 and Task 3 are of equal value. 3 In Task 2 and Task 3 there are several problems in (A) and (B). Problems at the beginning of (A) and the beginning of (B) are more important (in score) than problems at the end of (A) and the end of (B). 4 Errors and omissions will be penalized.

three structured executive tests (Wisconsin Card Sort, delayed response and self-ordered pointing). These results indicated that the early effects of ageing were apparent in standard neuropsychological tasks, but not on tasks of multiple subgoal planning (see also the section on multiple errand planning, below). Levine et al. (1998) compared the performance of younger and older groups on an adapted version of the SET. The adaptations included clarifying the task rules, testing rule recall, and making more explicit the nature of subtask prioritization by visually highlighting the importance placed on certain subtasks. There was no age difference in performance on this task. However, when education and NART-IQ were partialled out, there was a “mild age-related tendency towards nonstrategic performance” (Levine et al., 1998, p. 254). MacLeod (2001) investigated the difference in performance between young, middle-aged, and old participants on the revised SET, which is part of the Behavioural Assessment of Dysexecutive Syndrome battery of tasks (Wilson, Alderman, Burgess, Emslie, & Evans, 1996). There was no effect of


121

age group on SET performance, but this result should be interpreted cautiously because when using the scoring criteria from the test manual most participants scores were at ceiling. As indicated above, the SET does not require explicit plan formulation. Instead the performance measures indicate the efficiency of plan implementation and on-line plan monitoring. Kliegel, McDaniel, and Einstein (2000) altered the SET to include a prospective memory paradigm. In their modified SET, participants were required to explicitly formulate a plan of action to carry out the disparate task requirements. Then, following a delay, they had to execute the plan. In several studies, Kliegel and colleagues have applied this procedure to investigate the effects of age on plan formation, retention, initiation, and fidelity (match between formulated and executed plan). Although it has been repeatedly suggested that prospective memory (i.e., remembering to implement a set of intentions) should benefit from appropriate planning (Ellis, 1996; McDaniel & Einstein, 2000), empirical evidence on this issue is scarce. In their first study applying this paradigm, Kliegel et al. (2000) looked at the effects of adult ageing on: the elaborateness of plan formulation to carry out the SET; remembering to initiate the execution of the plan at a specified time; efficacy of SET performance (i.e., did participants attempt the first items of as many of the six subtasks as possible within the time limit?); and accuracy of plan implementation. Older adults made less elaborate plans, and were less likely to remember to initiate the plans when the appropriate cue occurred. There was no age effect on the fidelity of plan implementation: most participants showed relatively low correspondence between the explicit plan that was made and the way that they actually executed performance on the SET task. Also, older adults’ explicit plans were less elaborate than those of young participants, and so there were fewer plan elements to implement. Older adults attempted fewer of the six tasks in the SET than did younger adults within the time limit: this is an indicator of poor planning because to attempt fewer than six subtasks suggests that participants have not implemented the rule that the beginning of the tasks should be prioritized. Kliegel et al. interpret these results to indicate that there are considerable age decrements in complex prospective memory performance: i.e., the actual execution of the intention to work on all six tasks. In addition, these age decrements in task execution were highly correlated with the less elaborate plans of the older participants. Therefore, it seems plausible that appropriate and detailed planning by younger participants leads to better executed plans. In another study looking at explicit plan formulation in the SET, Kliegel, Martin, McDaniel, and Einstein (2002a) propose that planned intention formation is the first phase in a process model of complex prospective memory performance. In a study with 80 adults, Kliegel et al. (2002a) show that intention formation (as assessed by the elaborateness of an explicit plan for

122


carrying out the SET) correlates highly with other laboratory measures of executive functioning, particularly planning as assessed by a computerized “plan-a-day” task, in which participants have to run a number of errands in a fictitious setting. Plan formulation on the SET was not predicted by performance on the TOL planning task. Explicitly targeting the influence of good planning on age effects in complex prospective memory performance on the SET task, Kliegel et al. (2002a) experimentally manipulated the instructions given in the planning phase. In three experiments, they examined whether and how planning aids affect plan formulation and execution. In the first experiment, planning aids were provided which emphasized the need to move from one subtask to another in the SET. This resulted in improvements in task execution for both younger and older adults, but older adults still attempted fewer of the subtasks in the SET than young. The second experiment examined the effects of a planning aid that also targeted the initiation of the prospective action, and this benefited prospective memory initiation for both younger and older adults. The results showed that planning aids improved the elaborateness of the plans formulated by older adults but not those of younger adults, whose plans were elaborate without planning aids. With planning aids, the explicit plans made by older adults were as elaborate as those made by young, but even with improved explicit plans the older adults still attempted fewer of the subtasks of the SET. This suggests that even with much improved plan formulation, older adults were less able to execute an efficient plan on the SET. This is not simply due to forgetting the plan, because there were no age differences in plan retention. These results indicate that although planning aids may help older adults to formulate accurate plans, they may still have failures of plan implementation due to on-line working memory failures to execute intentions. A regression analysis suggested that age-related difficulties in plan implementation could be statistically explained in terms of working memory declines in older adults. The third experiment finally revealed that a combination of a specific planning aid and the provision of a visual planning schema that reduced working memory demands increased older adults’ performance on the SET task up to the level of younger adults.

AGE AND ERRAND PLANNING IN A NATURALISTIC SETTING The SET was devised by Shallice and Burgess (1991) in order to provide a better analogy to the complex scheduling activities of everyday life. In the same paper, they reported a task (the Multiple Errands Task, MET) in which planning must take place in a real shopping environment. Shallice and Burgess point out that it is only when patients with frontal lobe lesions are required to plan activities over longer time periods, or have to weigh up priorities in the face of competing demands, that their cognitive deficit


123

becomes apparent. The MET required planning of eight shopping-type errands in a real-life location within the constraints of task rules. Seven of the errands were very simple (e.g., buy a loaf of bread) while the eighth involved more difficult problem solving (e.g., requiring the participant to obtain four pieces of information and write them on a postcard). Performing these errands within a real-life context of unforeseen events makes the planning involved opportunistic and context dependent, i.e., the plan can be altered quite dramatically by the parameters of the setting, while social interaction enriches the realism of the test. Shallice and Burgess found that patients with frontal lobe damage made significantly more rule breaks than the control participants, and showed less efficient planning. Also the frontal lobe patients were more likely to become involved in social complications while performing the task. Garden et al. (2001) report performance on a more complex variation of the MET in younger and middle-aged adults. The task was adapted to make it suitable for the setting and population tested. Participants were given a list of 15 errands, ranging from very simple (buy any daily newspaper) to more complex errands that required problem solving and careful time scheduling. (Some information is urgently required. Buy a postcard and write on the postcard the average temperature in Edinburgh yesterday, and post it at the St. Nicholas postbox 20 minutes from now to [a specified name and address].) The task was designed so that it was very difficult to complete all of the errands within the time limit, especially if participants obeyed all of the task rules. The rules specified, for example, that errands could only be carried out in shops specified on a provided map, and that participants must not enter a shop without buying something. A range of measures were taken to assess the efficiency of the route taken, efficiency of planning, and the extent of rule breaking. There was no evidence of age differences in planning ability in terms of the route taken, or the scheduling of the errors. Although younger participants attempted more of the errands than middle-aged adults, the two groups did not differ on the number of errands that were completed. In relation to rule breaks, younger participants were more likely to break one of the task rules, i.e., to enter a shop during the task without buying anything. There was no evidence of any early age-related deficit in planning ability on the MET. As discussed above, middle-aged adults performed significantly worse than young on standard neuropsychological tests of executive functioning. It is possible that in the MET the participants may have been able to compensate for any executive deficit due to their acquired knowledge of shopping skills. This suggests an interesting dissociation in the effects of frontal lobe lesions and early ageing on abstract versus realistic planning tasks. Frontal lobe lesions often result in executive deficits which impair the ability to deal with real-world planning, but do not considerably affect

124


laboratory performance (McGeorge et al., 2001; Shallice & Burgess, 1991; Wilson et al., 1996). The opposite pattern appears to be the case in the course of normal ageing: although middle-aged adults perform poorly on structured laboratory neuropsychological tests, performance on more real-world planning tasks may be relatively intact. Real-world executive skills, such as the planning and monitoring involved in making a shopping trip, cooking a meal or dealing with a hectic work schedule, may be relatively robust in the face of early age-related neurological change. However, there have been few other ageing studies which have investigated the effects of ageing on realworld planning performance in complex but familiar environments. It would be of interest to see some more empirical studies of age differences in planning ability as applied to cooking or shopping or work-related scheduling tasks.

COMPARISON OF AGE EFFECTS ON REALISTIC AND ABSTRACT PLANNING TASKS In their overview of planning and control processes across the lifespan, Lachman and Burack (1993) state that there is evidence from controlled laboratory planning studies that older adults are not as efficient at planning as younger adults. However, they argue that familiarity of a planning task may influence performance motivation in old age and – as Chalmers and Lawrence (1993) have suggested – older adults’ planning abilities may be underestimated when task material does not reflect older adults’ experience in everyday planning situations. Therefore, Kliegel, Martin, McDaniel, and Phillips (2002c) conducted a study to experimentally examine, whether, how, and under which circumstances older adults are able to compensate for their cognitive deficits in the context of performing a planning task where the stimulus material varies in its resemblance to real-life experience. Kliegel et al. (2002c) report two experiments investigating parallel planning tasks that differ in whether the material is abstract or contextually meaningful. In both experiments, standard measures of cognitive processing resources were taken (speed, memory span, inhibition). In Experiment 1, 30 young and 30 older participants had to work on a contextual planning task based on Bisiacchi et al.’s (1998) errand planning task, whereas in Experiment 2 a modified version of the same planning task was administered to 30 young and 30 older adults that transferred the surface structure into an artificial environment, without changing the task, in principle. As an example, instead of withdrawing money from a bank in a certain street with specific opening hours of the bank counter (contextual task), participants had to get some gold from a certain planet with specific opening hours of the airport (artificial task). Moreover, in both experiments, the ability to focus on task-relevant information was assessed, in order to investigate whether there were age differences


125

in task-relevant prioritization of relevant material in the artificial and contextual tasks. Kliegel et al. (2002c) reported that in both experiments there were typical age-related declines in cognitive resource measures. In the case of the contextual task, there was no age deficit in planning, but in the artificial task (which had exactly the same structure and number of planning elements), there was a significant age decline in planning performance. In the contextual task, older adults showed better strategic selection of information than young (i.e., recalled proportionally more plan-relevant than plan-irrelevant information), whereas there was no age difference in strategic informational recall on the artificial planning task. In both planning tasks, resource variables (working memory, inhibition and speed) were all related to efficiency of planning performance. Applying multiple regression analyses, Kliegel et al. demonstrated that in the contextual planning task older adults were able to compensate for any resource-related deficits in planning performance by utilizing the selective strategy of focusing on relevant task features. This ability to compensate for resource deficits is not seen in the artificial planning task. This research suggests not only that older adults may be relatively good at planning in real-life settings, but gives some indication that this may be due to good ability to strategically inhibit irrelevant information where a setting is familiar and the information is contextually meaningful.

ADULT AGEING AND PLANNING: THEMES EMERGING FROM THE LITERATURE A summary of the age effects on the major planning tasks can be found in Table 6.1.

Realistic versus abstract planning tasks It has been recognized in the neuropsychological literature on planning that there is a crucial distinction between laboratory-based, highly constrained planning tasks, and planning carried out in real environments, which tends to be less structured (e.g., Shallice & Burgess, 1991; Wilson et al., 1996). Planning tasks set in complex environments are more sensitive to head injury than standard neuropsychological planning tasks (McGeorge et al., 2001). From the literature reviewed so far, it appears that older adults perform reliably poorly on abstract planning tasks, such as the TOL (e.g., Andrés & Van der Linden, 2000; Gilhooly et al., 1999; Robbins et al., 1998), yet perform relatively well at more familiar planning tasks, such planning shopping errands (Garden et al., 2001; Kliegel et al., 2002c). However, there are also a number of studies which indicate age differences in paper and pencil planning tasks that appear to be contextually familiar (e.g., errand and party planning:

126


TABLE 6.1 Summary of adult age effects on main measures of planning (see text for detailed findings and relevant references) Task

Measure of performance

TOL

Time to make mental plan Efficiency of mental plan

Number of moves to solution Plan implementation

Effects of age Old slower to make short plans but no age effect on complex TOL trials. Old poorer at estimating number of moves for solution and articulating mental plan. Old usually require more moves to solve trials. Old poorer match verbal plan and executed moves.

P+P plan formulation

Intrusion errors

SET

Plan formulation Plan implementation Efficiency of scheduling subtasks Breaking task rules

Old less elaborate plans. No age difference. Usually no age effect.

Scheduling and execution of errands Breaking task rules

No age effects.

Breaking task rules

MET plan execution

Old more likely to include less relevant items in plan. Old more likely to break rules.

Young more likely to break rules.

Young more likely to break rules.

Note: TOL = Tower of London. P+P = paper and pencil. SET = Six Elements Task. MET = Multiple Errands Task.

Bisiacchi et al., 1998; MacLeod, 2001). It is difficult to ascertain the extent to which these paper and pencil planning tasks can be considered realistic analogues of real-life planning skills. Also, despite older adults making poorer plans, they did tend to focus on the most salient aspects of the plan, whilst leaving out the finer details (MacLeod, 2001), and remember plan-salient information effectively (Martin & Ewert, 1997). It might therefore be the case that older adults were performing well on some aspects of planning in these tasks, in particular, selectively attending to and remembering the most critical plan-salient information. As outlined above, Kliegel et al. (2002c) made a direct investigation of the effects of increasing the familiarity of material on errand planning tasks, and found that older adults were able to take advantage of the familiar shopping context, resulting in no age differences in planning performance. These findings are consistent with the disuse hypothesis of ageing, which proposes that age-related declines in cognitive performance are largest on skills that have not been recently practised. These results have important implications. While there is considerable evidence that older adults show


127

poorer planning performance on abstract laboratory planning tasks, there is little evidence to suggest that their performance on real-life planning tasks is impaired. Much of the planning that we do in everyday life follows a daily routine (e.g., cooking), takes place in familiar environments (e.g., shopping), or has important implications for future well-being (e.g., financial planning). There have been very few studies of age effects on any of these planning abilities. The majority of studies do not look at planning in real environments because it is difficult to assess planning empirically while keeping the context realistic. However, this means that it may be inappropriate to extrapolate from findings of age impairment on planning tasks such as the TOL to implications for real-life planning. It is also likely that in real-life planning, external aids such as diaries and checklists will be used by older adults to compensate for any declines in memory. These findings in relation to planning also fit in with a recent meta-analysis of age effects on prospective memory tasks (Henry, MacLeod, Phillips, & Crawford, in press). Henry et al. report that there are strong age-related declines in performance on laboratory-based prospective memory tasks, but equally strong age-related improvements in performance on naturalistic prospective memory tasks. Age differences in planning therefore seem larger in tasks which are novel and abstract, and smaller in planning tasks which are more familiar and realistic. However, there have been very few age studies of truly realistic planning tasks. Furthermore, it is not clear to what extent paper and pencil tasks which use familiar contexts such as shopping errands or event planning should be considered as “realistic”. It would be useful to have more research into age effects on real-life planning skills, and the consequences that any age changes in planning have on functional impairment in old age. Thus, the first conclusion is that there are reliable age declines in performance on abstract, laboratory-based planning tasks. Some studies which have investigated age effects on realistic planning tasks indicate little age-related impairment, but this topic remains underexplored.

Are all planning tasks measuring similar cognitive functions? Planning is not likely to be a unitary function. We have outlined here a range of very different paradigms used to assess planning, and it is by no means clear that these different tasks assess parallel cognitive functions. Also, planning tasks are complex, and involve a number of different stages to achieve efficient performance. Most planning tasks measure either the explicit formulation of a plan (e.g., most paper and pencil errand planning tasks, the Virtual Planning Task), or the execution of a plan without the requirement for explicit plan formulation (e.g., TOL, SET, MET). There are relatively few

128


studies in which both the formulation and the subsequent execution of a plan are assessed. MacLeod (2001) looked at the relationship between a number of the planning tasks described earlier: the TOL, the party planning task, and the SET. There was no relationship between the number of moves made on the TOL task and any of the other planning measures. Those who took longer during the move phase of the TOL tended to make more errors in the party planning task. Performance on the party planning errand task was not associated with performance on the SET, or other planning subtests of the BADS battery (Behavioural Assessment of the Dysexecutive Syndrome, Wilson et al., 1996). Kliegel et al. (2002a) also found low correlations between SET performance and the TOL. This suggests that different planning tasks may require different cognitive processes for effective performance, and people who plan well on certain tasks may not plan well on other tasks. It may be the case that a range of planning tasks should be used to assess planning function before conclusions can be drawn about individual planning ability. Thus, a second conclusion is that it is a mistake to think of planning as a unitary function – different planning tasks appear to measure quite different aspects of cognitive functioning.

The role of working memory and executive functions in age differences in planning In an attempt better to understand the nature of age differences in planning, some studies have explored the role of working memory. There is evidence that in a population aged between 17 and 74 the number of excess moves made in the TOL task are related to measures of visuo-spatial working memory, but not to measures of verbal working memory (Gilhooly, Wynn, Phillips, Logie, & Della Sala, 2002). Also, statistical analysis indicates that age differences in TOL performance can be partially explained in terms of a visuospatial (but not verbal) working memory deficit (Phillips & Forshaw, 1998). Measures of working memory capacity have also been found to relate to performance on a holiday planning task (Martin & Ewert, 1997) and both abstract and contextual conditions of an errand planning task (Kliegel et al., 2002a). Working memory variance has been found to explain a proportion of the age differences in an errand planning task (Kliegel et al., 2002a) and the ability to retain and execute a specific plan in the SET (Kliegel et al., 2000). It would be interesting to see some further investigation of the role of working memory in age differences in planning, particularly using experimental manipulations which increase or decrease the working memory load of planning tasks to examine the effect of such manipulations on age differences. It has also been suggested that inhibitory functioning might be important in age differences in planning ability. Martin and Ewert (1997) found that


129

inhibitory efficiency and flexible use of attentional resources could account for much of the variance in performance of a complex holiday planning task in older adults. They argue that older adults have been found to have poor inhibitory control (Hasher & Zacks, 1989) and that inhibiting irrelevant stimuli is essential in effective planning in order to optimize efficient use of working memory (Harnishfeger, 1995). Martin and Ewert (1997) and Kliegel et al. (2002c) found that the ability to selectively remember planrelevant material (as opposed to peripheral information) was a good predictor of planning performance. This result indicates that good ability to selectively activate important information whilst inhibiting unimportant information is important in planning performance. Kliegel et al. argue that particularly where task materials are abstract, older adults have difficulty in inhibiting irrelevant information, and this may explain their planning difficulties. MacLeod (2001) found that the errors in the initial holiday plans made in the Virtual Planning Task (Miotto & Morris, 1998) did not correlate with any of the attentional measures in the Test of Everyday Attention (Robertson et al., 1994). In this planning task, participants make an initial plan, and then are given a chance to amend their plan if any errors are found. The number of errors after amendment of plans correlated with two selective attention measures. This suggests that the ability to selectively attend to information is important in the monitoring and adjustment of a plan. Thus, a third conclusion is that there is evidence which suggests that reduced working memory capacity may underlie age differences in formulating mental plans, and also accurately retaining and executing plans. When faced with a novel task, selectively attending to important information and inhibiting irrelevant information may become less efficient with age, and this might explain age-related inefficiencies in scheduling multiple plan elements and formulating plans.

Optimizing planning performance in older adults There are some suggestions in the literature on how to improve planning performance in older adults. Chalmers and Lawrence (1993) investigated the effects of planning aids on adults’ organization of a complex party planning task. Their results indicated that older adults’ performance was worse compared to younger adults’, but that planning aids designed to encourage subjects to organize their plans according to specific characteristics improved planning performance in older adults. Kliegel et al. (2002b) also found that planning cues that targeted specific aspects of task rules and task initiation in the SET resulted in substantially more elaborate plans in older adults. It would be interesting to investigate whether specific interventions designed to reduce the working memory load of planning tasks and the tendency

130


to attend to irrelevant aspects of the task might improve older adults’ planning. Martin and Ewert (1997) argue that in the complex environment of everyday life there is a constant demand for people to make, execute and monitor plans, resulting in frequent planning practice. This suggests that everyday practice in carrying out tasks such as shopping, or planning finances or holidays, is likely to protect older adults against the deleterious effects of cognitive changes with age. This leads to a fourth conclusion, that there are some promising suggestions that older adults’ plan formulation and execution can be improved using cueing techniques.

Frontal lobe involvement in age-related planning impairment Although much recent research into cognitive ageing has focused on the role of the frontal lobes of the brain, it is agreed that other brain regions also show age-related change (e.g., Greenwood, 2000). However, there have to date been no published neuroimaging studies that directly investigate the involvement of particular brain areas in age-related change in planning. Nevertheless, the pattern of performance of older adults and patients with frontal lobe injuries on planning can be compared. Declines in working memory capacity (Gilhooly et al., 1999), as well as inhibitory and attentional functions (Martin & Ewert, 1997), may contribute to the poorer planning performance of older adults. This might seem to lend support to the idea that poorer planning performance in older adults may be related to frontal lobe declines (West, 1996). While there is clear evidence that both frontal lobe patients and older adults are impaired at some of the same planning tasks, the two groups may be failing the tasks for different reasons. For example, the breaking of social rules and conventions by patients with gross frontal lobe lesions may impair their ability to execute an efficient plan in a complex real environment (Goldstein, Bernard, Fenwick, Burgess, & McNeil, 1993; Shallice & Burgess, 1991), but older adults do not tend to show such disorders of social behaviour (McPherson, Phillips, & Della Sala, 2002). Also, older adults seem to show particular deficits in abstract planning tasks compared to real-world tasks (e.g., Kliegel et al., 2002c) whereas frontal lobe patients may do relatively well on some laboratory planning tasks but show planning deficits in real life (e.g., Shallice & Burgess, 1991). On tasks in which both older adults and frontal lobe patients show deficits, there may be different reasons for the planning difficulties. For example, on the TOL task, it has been argued that frontal lobe involvement in the task is not attributable to working memory deficits, and that instead the problem lies


131

in dealing with goal conflicts. In contrast, for older adults there is some evidence that reduced working memory capacity may underlie poor performance on the TOL. Robbins et al. (1998) argue that the pattern of age-related decline in performance on tasks such as TOL does not resemble the pattern seen in frontal lobe patients, and instead parallels more closely the cognitive pattern seen following basal ganglia damage. Further, Crawford et al. (2000) argue that age differences in a range of executive tasks, including the TOL, are not evidence of a differential deficit in “frontal lobe functioning” because all of the age variance in the executive tasks can be explained by performance on general cognitive ability tests. Thus, a final conclusion is that although frontal lobe patients and older adults show deficits on planning tasks, there is evidence to suggest that these deficits may occur for different reasons, and therefore age differences in planning should not routinely be interpreted as indicating “frontal lobe” deficits.

ACKNOWLEDGEMENT We would like to acknowledge the assistance of Barbara MacLeod in preparing this chapter.

REFERENCES Allmano, N., Della Sala, S., Laiacona, M., Pasetti, C., & Spinnler, H. (1987). Problem solving ability in aging and dementia: normative data on a non-verbal test. Italian Journal of Neurological Sciences, 8, 111–120. Andrés, A. U., & Van der Linden, M. (2000). Age-related differences in supervisory attentional functions. Journals of Gerontology Series B – Psychological Sciences and Social Sciences, 55, 373–380. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. Bower (Ed.), Recent advances in learning and motivation, VIII (pp. 47–90). New York: Academic Press. Baker, S. C., Rogers, R. D., Owen, A. M., Frith, C. D., Dolan, R. J., Frackowiak, R. S. J., & Robbins, T. W. (1996). Neural systems engaged by planning – a PET study of the Tower of London task. Neuropsychologia, 34, 515–526. Bisiacchi, P. S., Sgaramella, T. M., & Farinello, C. (1998). Planning strategies and control mechanisms: Evidence from closed head injury and aging. Brain and Cognition, 37, 113–116. Chalmers, D., & Lawrence, J. A. (1993). Investigating the effect of planning aids on adults and adolescents organisation of a complex task. International Journal of Behavioral Development, 16, 191–214. Connor, L.T., Dunlowsky, J., & Hertzog, C. (1997). Age-related differences in absolute but not relative metamemory accuracy. Psychology and Aging, 12, 50–71. Crawford, J. R., Bryan, J., Luszcz, M. A., Obonsawin, M. C., & Stewart, L. (2000). The executive decline hypothesis of cognitive aging: Do executive deficits qualify as differential deficits and do they mediate age-related memory decline? Aging, Neuropsychology and Cognition, 7, 9–31. Ellis, J. A. (1996). Prospective memory or the realization of delayed intentions: A conceptual framework for research. In M. Brandimonte, G. O. Einstein, & M. A. McDaniel (Eds.),

132


Prospective memory: Theory and applications (pp. 1–22). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Garden, S. E., Phillips, L. H., & McPherson, S. E. (2001). Mid-life aging, open-ended planning and laboratory measures of executive function. Neuropsychology, 15, 472–482. Gilhooly, K. J., Phillips, L. H., Wynn, V. E., Logie, R. H., & Della Sala, S. (1999). Planning processes and age in the 5-disk Tower of London task. Thinking and Reasoning, 5, 339–361. Gilhooly, K. J., Wynn, V. E., Phillips, L. H., Logie, R. H., & Della Sala, S. (2002). Visuo-spatial and verbal short term/working memory in the five-disk Tower of London task: An individual–differences approach. Thinking and Reasoning, 8, 165–178. Goldstein, L. H., Bernard, S., Fenwick, P. B. C., Burgess, P. W., & McNeil, J. (1993). Unilateral frontal lobectomy can produce strategy application disorder. Journal of Neurology, Neurosurgery and Psychiatry, 56, 274–276. Greenwood, P. M. (2000). The frontal aging hypothesis evaluated. Journal of the International Society, 6, 705–726. Gur, R. C., Gur, R. E., Obrist, W. D., Skolnick, B. E., & Reivich, M. (1987). Age and regional cerebral blood flow at rest and during cognitive activity. Archives of General Psychiatry, 44, 617–621. Harnishfeger, K. K. (1995). The development of cognitive inhibition: Theories, definitions and research evidence. In F. N. Dempster & C. J. Brainerd (Eds.), Interference and inhibition in cognition (pp. 175–204). San Diego, CA: Academic Press. Hasher, L., & Zacks, R. T. (1988). Working memory, comprehension, and aging: A review and a new view. In G. H. Bower (Ed.), The psychology of learning and motivation (pp. 193–225). San Diego, CA: Academic Press. Hayes-Roth, B., & Hayes-Roth, F. (1979). A cognitive model of planning, Cognitive Science, 3, 275–310. Henry, J. D., MacLeod, M. S., Phillips, L. H., & Crawford, J. R. (in press). A meta-analytic review of prospective memory and aging. Psychology and Aging. Kliegel, M., McDaniel, M. A., & Einstein, G. O. (2000). Plan formation, retention, and execution in prospective memory: A new approach and age-related effects. Memory & Cognition, 28, 1041–1049. Kliegel, M., Martin, M., McDaniel, M. A., & Einstein, G. O. (2002a). Complex prospective memory and executive control of working memory: A process model. Psychologische Beiträge, 44, 303–318. Kliegel, M., Martin, M., McDaniel, M. A., & Einstein, G. O. (2002b). Prospective memory and aging: How planning affects performance. Manuscript submitted for publication. Kliegel, M., Martin, M., McDaniel, M. A., & Phillips, L. H. (2002c). Older adults can compensate for cognitive deficits in planning performance – but everyday experience matters. Manuscript submitted for publication. Kray, J., & Lindenberger, U. (2000). Adult age differences in task switching. Psychology & Aging, 15, 126–147. Lachman, M. E., & Burack, O. R. (1993). Planning and control processes across the life span: An overview. International Journal of Behavioral Development, 16, 131–143. Levine, B., Stuss, D. T., Milberg, W. P., Alexander, M. P., Schwartz, M. F., & MacDonald, R. (1998). The effects of focal and diffuse brain damage on strategy application: Evidence from focal lesions, traumatic brain injury and normal aging. Journal of the International Neuropsychological Society, 4, 247–264. McDaniel, M. A., & Einstein, G. O. (2000). Strategic and automatic processes in prospective memory retrieval: A multiprocess framework. Applied Cognitive Psychology, 14, S127–S144. McGeorge, P., Phillips, L. H., Crawford, J. R., Garden, S. E., Della Sala, S., Milne, A. B., Hamilton, S., & Callender, J. S. (2001). Using virtual environments in the assessment of executive dysfunction. Presence: Teleoperators and Virtual Environments, 10, 375–383.


133

MacLeod, M. S. (2001). Cognitive ageing and the role of the frontal lobes in prospective memory and planning. Unpublished doctoral dissertation, University of Aberdeen. McPherson, S. E., Phillips, L. H., & Della Sala, S. (2002) Age, executive function and social decision-making: A dorsolateral prefrontal theory of cognitive aging. Psychology and Aging, 17, 598–609. Martin, M., & Ewert, O. (1997). Attention and planning in older adults. International Journal of Behavioral Development, 20, 577–594. Miotto, E. C., & Morris, R. G. (1998). Virtual planning in patients with frontal lobe lesions. Cortex, 34, 639–657. Morris, R. G., Miotto, E. C., & Feigenbaum, J. D. (1997). Planning ability after frontal and temporal lobe lesions in humans: The effects of selection equivocation and working memory load. Cognitive Neuropsychology, 14, 1007–1027. Nielson, K. A., Langenecker, S. A., & Garavan, H. (2002). Differences in the functional neuroanatomy of inhibitory control across the adult life span. Psychology & Aging, 17, 56–71. Norman, D. A., & Shallice, T. (1986). Attention to action: Willed and automatic control of behaviour. In R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.), Consciousness and selfregulation, advances in research and theory (Volume 4, pp. 1–18). New York: Plenum Press. Owen, A. M. (1997) Cognitive planning in humans: Neuropsychological, neuroanatomical and neuropharmacological perspectives. Progress in Neurobiology, 53, 431–450. Owen, A. M., Doyon, J., Petrides, M., & Evans, A. C. (1996). Planning and spatial working memory: A positron emission tomography study in humans. European Journal of Neuroscience, 8, 353–364. Pentland, L., Todd, J. A., & Anderson, V. (1998). The impact of head injury severity on planning ability in adolescence: A functional analysis. Neuropsychological Rehabilitation, 8, 301–317. Phillips, L. H., & Forshaw, M. J. (1998). The role of working memory in age differences in reasoning. In R. H. Logie & K. J. Gilhooly (Eds.), Working memory and thinking (pp. 23–43). Hove, UK: Psychology Press. Phillips, L. H., Gilhooly, K. J., Logie, R. H., Della Sala, S., & Wynn, V. E. (2003). Age, working memory, and the Tower of London task. European Journal of Cognitive Psychology, 15, 291–312. Phillips, L. H., McPherson, S. E., & Della Sala, S. (2002). Age, cognition and emotion: The role of anatomical segregation in the frontal lobes. In J. Grafman (Ed.), Handbook of neuropsychology. Volume 7: The frontal lobes (2nd ed., pp. 73–97). Amsterdam: Elsevier Science BV. Phillips, L. H., Smith, L., & Gilhooly, K. J. (2002). The effects of age and induced positive and negative mood on planning. Emotion, 2, 263–272. Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology A, 54, 579–598. Raz, N. (1996). Neuroanatomy of the aging brain observed in vivo: A review of structural MRI findings. In E. D. Bigler (Ed.), Neuroimaging II: Clinical applications (pp. 153–184). New York: Plenum Press. Robbins, T. W., James, M., Owen, A. M., Sahakian, B. J., Lawrence, A. D., McInnes, L., & Rabbitt, P. M. A. (1998). A study of performance on tests from the CANTAB battery sensitive to frontal lobe dysfunction in a large sample of normal volunteers: Implications for theories of executive functioning and cognitive aging. Journal of the International Neuropsychological Society, 4, 474–490. Robertson, I. H., Ward, T., Ridgeway, V., & Nimmo-Smith, I. (1994) The test of everyday attention. Bury St Edmunds: Thames Valley Test Company. Shallice, T. (1982). Specific impairments of planning. Philosophical Transactions of the Royal Society of London, B298, 199–209.

134


Shallice, T., & Burgess, P. W. (1991). Deficits in strategy application following frontal lobe damage in man. Brain, 114, 727–741. Welsh, M. C., Satterlee-Cartmell, T., & Stine, M. (1999). Towers of Hanoi and London: Contribution of working memory and inhibition to performance. Brain and Cognition, 41, 231–242. West, R. L. (1996). An application of prefrontal cortex function theory to cognitive aging. Psychological Bulletin, 120, 272–292. Wilson, B. A., Alderman, N., Burgess, P. W., Emslie, H., & Evans, J. J. (1996). Behavioural assessment of the dysexecutive syndrome. Bury St Edmunds: Thames Valley Test Company.

CHAPTER SEVEN

Cognitive planning in humans: New insights from the Tower of London (TOL) task Adrian M. Owen MRC Cognition and Brain Sciences Unit, Cambridge, UK

INTRODUCTION The frontal lobes have long been thought to play an important role in planning behaviour. For example, Harlow (1868), argued that frontal lobe lesions in humans result in a loss of “planning skill”, whilst much later Bianchi (1922) described a loss in the ability to “coordinate the different elements of a complex activity” in monkeys with large frontal lesions. More contemporary accounts have characterized the role of the frontal cortex in planning behaviour using similarly descriptive, terms; e.g., “as a general system for sequencing or guiding behaviour towards the attainment of an immediate or distant goal” (Jouandet and Gazzaniga, 1979), or as crucial for the “planning of future actions” (for review, see Shallice, 1988). Until recently, however, the assumed relationship between cognitive planning and the frontal lobes lacked solid empirical support and was based largely on anecdotal reports of disorganized behaviour in patients with relatively non-specific brain injury, or on the behaviour of monkeys with large excisions of the frontal cortex. Moreover, planning difficulties are not unique to patients with circumscribed frontal lobe damage. For example, “frontal-like” planning deficits have been described in patients with mild Parkinson’s disease (Morris, Downes, Evenden, Sahakian, Heald, & Robbins 1988; Owen et al., 1992; Owen, Sahakian, Hodges, Summers, Polkey, & Robbins, 1995a; Owen, Doyon, Dagher, & Evans, 1998), and other basal-ganglia disorders, suggesting that an equivalence between the prefrontal cortex and planning function cannot be assumed. 135

136

OWEN

In recent years, a substantial number of studies have elected to use versions of the TOL task to investigate the relationship between cognitive planning and brain function in humans (e.g., Baker et al., 1996; Dagher, Owen, & Brooks, 1999; Dagher, Owen, Boecker, & Brooks, 2001; Hodgson, Tiesman, Owen, & Kennard, 2002; Morris et al., 1988; Morris, Ahmed, Syed, & Toone, 1993; Owen, Downes, Sahakian, Polkey, & Robbins, 1990; Owen, et al., 1992; Owen et al., 1995a; Owen, Doyon, Petrides, & Evans, 1996a; Owen, et al., 1998; Rowe, Owen, Johnsrude, & Passingham, 2001; Shallice, 1982). As a direct result, a wealth of new evidence has accumulated, both supporting a role for the frontal lobe in planning behaviour and providing more precise psychological information about the nature of that role. This chapter will review that evidence and in doing so, attempt to summarize the findings in terms of their contribution to current understanding about the cognitive and neuroanatomical bases of complex planning behaviour.

THE TOWER OF LONDON (TOL) The TOL (see chapter 1 by Ward & Morris, Figures 1.1 and 1.2) clearly requires “forward thinking” or planning, since an early, incorrect move can render the problem virtually unsolvable unless all previous steps are retraced and reversed, in order to correct the inappropriate move. In the original study, reported in Shallice (1982), patients with left anterior cortical pathology were shown to be impaired in terms of the number of moves required to complete the problems. This important finding has been followed up by a number of different studies that have used a computerized version of the TOL task, presented on a touch-sensitive screen (e.g., Morris et al., 1988; Owen et al., 1990, 1992, 1995a). In many of the computerized versions of the task, the subjects are presented with the starting position in the bottom of screen and the goal position in the top (Figure 7.1). The bottom display is rearranged, and this is facilitated using a touch-sensitive screen. In a version used in the studies presented below, instead of balls or discs threaded on rods, there are balls threaded into socks, the so-called Stockings of Cambridge. As with the original Shallice (1982) version, the position of the balls in the top half of the screen is varied for each problem such that a solution can be reached in a minimum of two, three, four or five moves. In general, the two-move TOL problems require very little cognitive planning, and can be solved using a simple, visual matching-to-sample strategy. This strategy, whereby each ball in the bottom half of the screen is moved directly to its goal position (i.e., the position corresponding to the same coloured ball in the top half of the screen), is illustrated at the top of Figure 7.1 (see Figure 7.1(a)). In contrast, the more difficult three-, four- and fivemove problems can often not be solved using a visual matching-to-sample

Figure 7.1 Two trials from the computerized version of CANTAB TOL task (Owen et al., 1990). The problem shown at (a) requires two moves. A more difficult four-move problem is also shown (b). Participants are required to move the set of balls around in the bottom half of each configuration to match the goal arrangement in the top half. A ball can be moved by first touching it and then touching one of the empty positions in the array.

138

OWEN

strategy. In fact, in many cases these problems require the subject to make visually counter-intuitive moves (i.e., to move a ball away from its final destination), in order to execute the appropriate solution, which involves a considerable amount of “thinking ahead” or planning. This more complex task requirement is illustrated at the bottom of Figure 7.1 (Figure 7.1(b)). For each test problem, a “yoked” control condition is also given to provide a baseline measure of motor initiation and execution times. By subtracting the response time from each move in this control condition from that of the corresponding move in the planning condition, estimates of “thinking time” are generated. Initial thinking time refers to the time between the presentation of the planning problem and the first touch, minus the corresponding motor initiation time from the yoked control condition. Subsequent thinking time refers to the time between the selection of the first ball and the completion of the problem minus the total motor execution time derived from the corresponding control problem. Owen et al. (1990) assessed performance on this task in 26 neurosurgical patients with unilateral or bilateral frontal lobe excisions and later (Owen, Sahakian, Semple, Polkey, & Robbins, 1995b), in a group of 20 patients with unilateral temporal lobe excisions and a group of 11 patients in whom the more selective amygdalo-hippocampectomy had been performed. Compared to age and IQ matched controls, the frontal lobe group required more moves to complete the problems and produced fewer perfect solutions. Initial “thinking”, or “planning” time was unimpaired in these patients although the amount of time spent thinking on-line (i.e., subsequent to the first move) was significantly prolonged. This pattern of impairment appears to be relatively specific at the cortical level since no deficits were observed in the two groups of neurosurgical patients with damage to the medial temporal lobe region (Owen et al., 1995b). In a follow-up study (Owen et al., 1995a), the TOL task was modified to examine the relationship between thinking (planning) time, problem difficulty and solution accuracy in the group of patients with frontal lobe excisions. Subjects were required to study each of the original TOL problems and then decide how many moves would be required to reach an ideal solution (i.e., with the minimum number of moves), without actually moving any of the balls. Because this modification required subjects to evaluate and solve the problems in full, without executing any of the necessary subgoals (i.e., moving the balls), it was no longer possible to compromise “initial planning time” (i.e., the time before a response was made) in favour of “on-line” consideration of the problem during the execution of the solution (i.e., “subsequent thinking time”). This “one-touch” modification to the task served to encourage subjects to plan the solution in full, before they initiated a response.

7. COGNITIVE PLANNING IN HUMANS

139

The results of the previous study were essentially confirmed: compared to the matched control group, the frontal lobe patients were significantly impaired in terms of solution accuracy, whilst solution latency (or “initial thinking time”) was relatively preserved. In summary, the results of these studies demonstrate a significant association between cognitive planning and the frontal cortex in humans, although they leave a number of important psychological and neuroanatomical issues unresolved. For example, like many other tests of planning ability, the TOL task is complex and the precise cognitive components that combine to produce an efficient plan of action have not been clearly defined experimentally. Moreover, while the results described above clearly implicate the frontal cortex in TOL performance, patient studies provide very little anatomical information about the precise frontal lobe regions involved, since the excisions are rarely confined to one or even a few cytoarchitectonic areas. In the remainder of this chapter, these two issues will be considered in detail in the light of new evidence which has emerged from behavioural studies of eye tracking during the TOL problems and from functional neuroimaging studies in healthy volunteers performing the same task.

PSYCHOLOGICAL CONSIDERATIONS: COGNITIVE COMPONENTS OF PERFORMANCE IDENTIFIED FROM EYE-TRACKING BEHAVIOUR Successful performance on the TOL task typically involves a number of stages. The overall situation is considered by assessing the initial and goal states with reference to differences in the positions and overall configuration of the balls. A series of subgoals is defined. A sequence of moves is generated to attain these subgoals. This sequence is refined and revised according to the results of mental rehearsal and, finally, the correct solution is executed. A number of computerized models of problem solving have been constructed, which are capable of performing the TOL task (e.g., Anderson, 1993; Dehaene & Changeux, 1997; Newell & Simon, 1972). One particularly successful model is the SOAR production system (Newell, 1990). SOAR has no difficulty in maintaining current goals, the current problem state and other task relevant information during problem solving. However, the model does run into difficulty when there is no clear choice as to which move is the most beneficial in a given context. Under these conditions, the program creates a temporary subgoal to resolve the conflict. This situation arises during the TOL task when moving one of the balls directly into its target location prevents additional moves that are necessary to complete the entire problem. This complex task requirement is illustrated at the bottom of Figure 7.1 above (Figure 7.1(b)) and the only correct solution requires a “shunting” manoeuvre by which the ball is placed in a temporary (i.e., subgoal) location.

140

OWEN

The way in which models like SOAR might be implemented within biological systems remains to be specified. Several recent attempts to address this issue have focused on detailed analyses of behavioural strategies during performance of the TOL task (e.g., hand and eye movements). In one study (Hodgson et al., 2000), the pattern of natural scanning eye movements was examined while subjects performed the “one-touch” version of the TOL task described above (Owen et al., 1995a). Subjects were instructed to plan but not to execute the problem solutions, pressing a mouse key and responding verbally once the minimum number of moves required to reach a solution had been identified. To facilitate analysis of the eye-movement data, only two balls and problems requiring one, two or three moves were used. Eye movements were recorded using a video-based pupil tracker and both saccades and fixations were identified using a semi-automatic algorithm. Unsurprisingly perhaps, the total fixation time on the set of balls to be rearranged increased with the complexity of the problem. In contrast, however, the total time spent looking at the “target” set of balls remained constant despite the difficulty of the problem. Moreover, the proportion of time spent fixating the two halves of the display changed over the course of a trial; specifically (for the most difficult three-move planning problems), a brief initial period spent inspecting the goal arrangement was followed by a protracted period of saccades around the set of balls to be rearranged. Only at the end of each trial did the focus return once again to the target arrangement. One interpretation of these results is that the volunteers used direct fixation to acquire task-relevant information from the target arrangement and then held this information in memory during the elaboration of the solution (Hayhoe, Bensinger, & Ballard, 1997; Land & Furneaux, 1997). An alternative explanation is that the target arrangement is constantly monitored in parafoveal vision via covert attention, even when it is not being fixated directly. Hodgson et al. (2000) were unable to distinguish between these alternatives, although in related studies, block copying tasks have been used to demonstrate that covert attention is rarely used in this manner (Hayhoe et al., 1997). The finding that return fixations to the target arrangement occurred at the end of trials also suggests that subjects did not rely exclusively on either parafoveal vision or memory of the goal arrangement. Instead, this behaviour suggests that subjects refixated the goal at the end of each trial to confirm that the problem had been correctly solved. Aside from vertical shifts in gaze direction between the two visual fields, horizontal transformations in fixation between locations were also observed. These occurred more frequently in the set of balls to be rearranged than in the target set, suggesting that they correspond in some way to the planning of problem solutions. Although it was not possible to establish a one-to-one correspondence between these lateral gaze shifts and the ball movements


141

required to solve the problems, it seems most likely that they reflect visuomotor imagery during the elaboration of problem solutions (Brandt & Stark, 1997). These novel eye-movement findings suggest, therefore, that TOL performance involves the coordination of an appropriate gaze-shifting strategy. The nature of this strategy suggests that complex TOL problems proceed in several discrete phases. An initial assessment of the problem is followed by a solution elaboration phase in which different operations are rehearsed and assessed. Finally, there is a verification phase, during which an internal representation of the planned solution is compared to the goal configuration. A second, follow-up experiment was designed to test whether the distribution of gaze within individual TOL problems could be shown to be problem dependent. This was achieved by keeping the position of the balls to be rearranged constant from problem to problem and manipulating only the arrangement of balls in the target set. In this way, differences between problems in terms of the distribution of gaze on the set to be rearranged could be attributed to the planning of problem solutions, rather than simple differences in the arrangement of objects in the display. In addition, two types of problem were generated, termed “blue ball” problems and “non-blue ball” problems. Critically, blue ball problems required the shunting manoeuvre described above, whereby one ball (in all cases the blue ball) initially had to be moved away from the ball’s final destination to a temporary subgoal location to make room for other balls, before being finally moved into position (see Figure 7.1(b)). Blue ball problems could not be solved successfully unless the volunteer realized the importance of this manoeuvre. Efficient planners (those making fewer than 5% errors), biased their gaze disproportionately towards the blue ball location during blue ball problems, while inefficient planners (those making more than 5% errors) did not. In addition, the inclusion of “dummy” balls, which were irrelevant to the solution of the problem, confirmed that efficient planners biased their gaze away from these balls towards alternatives that were central to the final solution. These findings demonstrate that the distribution of gaze was sensitive to the particular ball moves being assessed by the subject. Efficient planners biased their gaze towards the blue ball location while planning solutions to blue ball problems and to the correct (non-dummy) ball locations on non-blue ball problems, even though the actual arrangement of balls being rearranged was identical on both types of trial. In summary, it is often assumed that advanced planning involves the construction of an internal program for the movement sequence, which is later recalled to control execution of the correct solution. The eye-movement data described above suggests that this “plan-as-program” view (Clark, 1997) may be incorrect. For example, no evidence was found for stereotypical sequences of eye movements, which might correspond to the rehearsal of a

142

OWEN

fully formed action sequence. In contrast, the fact that efficient planning involves selectively biasing gaze towards the problem-critical balls suggests an alternative mechanism through which advanced planning might benefit problem solving; that is, that directing gaze strategically during planning establishes a parsimonious motoric representation of the key features of the problem and facilitates refixation of the same areas of the display during problem solution.

ANATOMICAL CONSIDERATIONS: LOCALIZING THE CORE NEURAL SUBSTRATES OF PERFORMANCE USING FUNCTIONAL NEUROIMAGING The frontal cortex is not a homogeneous region of the brain, but comprises several architectonic areas that differ in terms of their connections with other cortical and subcortical areas (Pandya & Barnes, 1987). Relative to the enormous amount of information that is available about the structural and functional organization of the monkey brain, very little is known of the connections between specific cortical areas in humans. In spite of this, a comprehensive reparcellation and comparative cytoarchitectonic analysis of the human and macaque frontal cortex has revealed a remarkable degree of topographic and architectural similarity between the two species in this region (Petrides & Pandya, 1994). In patient studies, it is not possible to establish which areas of the frontal cortex are involved in a given cognitive process with any degree of anatomical precision since the excisions are rarely confined to specific cytoarchitectonic areas. In recent years, functional neuroimaging techniques such as single photon emission tomography (SPECT), positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), have provided a unique opportunity for assessing the relationship between patterns of cortical and subcortical activation and different aspects of cognitive planning in healthy control volunteers. Two early SPECT studies of planning in normal subjects demonstrated increased cerebral blood flow (CBF) in the frontal cortex during versions of the TOL task (Morris et al., 1993; Rezai, Andreasen, Allinger, Cohen, Swayze, & O’Leary, 1993). However, the spatial resolution of SPECT is not sufficient for investigating functional specialization within the human frontal cortex. In a later study (Owen et al., 1996a) PET, which has better spatial resolution, was used to examine regional CBF while subjects solved either simple or difficult TOL problems. Blood flow during these conditions was compared to that during a control condition, which involved identical stimuli and responses but required minimal planning. When activation in the control condition was subtracted from that during the difficult planning condition, a significant regional CBF change was observed in the mid-dorsolateral frontal cortex


143

(Figure 7.2(b)). In the human brain, this region comprises mainly cytoarchitectonic areas 9 and 46 which occupy the mid-part of the superior and middle frontal gyri, a considerable proportion of this cortex lying within the depths of the middle frontal sulcus (Figure 7.2(a)). Similar results were reported in a later study (Baker et al., 1996), which employed the modified “one-touch” version of the TOL task used by Owen et al. (1995a), to study patients. One significant problem with many of these studies is that the selection of a control task invariably determines, to a large extent, the pattern of activation observed. For example, as all cognitive tasks involve some planning, albeit at different levels of complexity, the relationship between the experimental (e.g., planning) task and the (frequently inadequately defined) planning demands of the control task can complicate the interpretation of imaging data. In addition, the visuomotor demands of the experimental (e.g., planning) and control tasks may differ even subtly, which may further complicate the interpretation of the results. One approach to this problem is to use a parametric or correlational task design, which involves no control task per se, but rather multiple scans with similar planning requirements, but different levels of task difficulty. Dagher et al. (1999) used this approach to examine regional cerebral blood flow, using PET, during increasingly complex TOL problems. Volunteers were scanned while performing one-, two-, three-, four- and five-move TOL problems and a rest condition which involved no task. Activity in the dorsolateral frontal cortex was shown to be complexity dependent, while activity in posterior parietal cortex and in the occipital lobe was shown to be complexity independent. This suggests that, while the dorsolateral frontal cortex plays a central role in planning solutions to the TOL problems, posterior cortical areas such as occipital and parietal cortex make more basic contributions to aspects of visual and spatial processing during the task. By correlating regional CBF changes with the number of moves made to reach a solution (irrespective of the minimum number of moves actually required to solve the problem), it was also possible to differentiate between regions involved in planning and those involved in movement execution. Within the basal ganglia, for example, movement-related changes were observed in the putamen, while problem-complexity-related (but movement unrelated) changes were observed in the caudate nucleus. The latter finding may help to explain why “frontal-like” TOL impairments are often observed in patient groups with basal ganglia pathology, such as Parkinson’s disease (e.g., Morris et al., 1988; Owen et al., 1992) and concur fully with the observation that task performance is accompanied by abnormal regional CBF changes in the basal ganglia in these groups (e.g., Owen et al., 1998; Cools, Stefanova, Barker, Robbins, & Owen, 2002). In summary, recent functional neuroimaging studies have been able to confirm and extend previous investigations in patients by identifying, more

Figure 7.2 (a) Schematic drawing of the lateral surface of the human brain to indicate the location of the dorsolateral frontal cortex (areas 9, 46 and 9/46). Adapted from Petrides and Pandya (1994). Numbers refer to cytoarchitectonic areas as defined by Petrides and Pandya (1994), and based on the original analysis by Brodmann (1908). (b) Average PET subtraction images shown superimposed upon the corresponding averaged MRI scan for the healthy control subjects included in the study by Owen et al. (1996a). The schematic (top right), illustrates the approximate position of the coronal slice shown in the lower half of the figure. Subtraction of a control condition from a TOL planning condition yielded the focal changes in blood flow shown as a t-statistic image, whose range is coded by the colour scale placed to the left of the figure. The sagittal section (bottom) illustrates the significant rCBF increase observed in the mid-dorsolateral frontal cortex. The coronal section at y = +35 also illustrates the rCBF increase in the right mid-dorsolateral frontal cortex, which just missed statistical significance.


145

precisely, the frontal cortical area that appears to be critical for performance on the TOL planning task: namely, the mid-dorsolateral frontal region. The combined evidence from different functional neuroimaging studies (e.g., Baker et al., 1996; Dagher et al., 1999; Owen et al., 1996a; 1998), together with previous investigations in patients (e.g., Owen et al., 1990; 1995a), suggests that within the dorsolateral frontal region neither hemisphere plays a disproportionate role, at least in the type of high-level planning that is required in the TOL task (but see Grafman and colleagues, chapter 6, this volume). It is important to emphasize, however, that none of the results discussed above suggest that this region of the dorsolateral frontal cortex is either wholly or uniquely involved in mediating cognitive planning processes. In the monkey it has been shown that specific regions of the lateral frontal cortex are reciprocally connected with multiple posterior cortical and sub-cortical regions, which undoubtedly reflects close functional relationships between anatomically distant areas. For example, the mid-dorsolateral region of frontal cortex (Brodmann areas 9 and 46; Brodmann, 1908) which appears to be most critical for cognitive planning is closely connected with the ventrolateral frontal cortex (Barbas & Pandya, 1989; Watanabe-Sawaguchi, Kubota, & Arikuni, 1991), and, at the same time, with the limbic region of the medial temporal lobe (Adey & Meyer, 1952; Nauta, 1964; Goldman-Rakic, Selemon, & Schwartz, 1984). In addition, descending cortico-striatal inputs to the caudate and the putamen project back to discrete frontal lobe regions, including the mid-dorsolateral frontal cortex (Middleton & Strick, 1994, 1995), via various thalamus nuclei, closing the so-called “cortico-striatal loops”. These neuroanatomical data concur fully with the results of the neuroimaging studies described above. Thus, a number of non-prefrontal cortical and sub-cortical regions were also activated by the versions of the TOL task used by Owen et al. (1996a) and/or Dagher et al. (1999), including the caudate nucleus, the presupplementary motor area, the anterior premotor cortex and the posterior parietal cortex. The available anatomical and functional neuroimaging data suggest, therefore, that whilst the mid-dorsolateral frontal cortex plays a critical role in complex planning behaviour, it does so through close functional interactions with multiple cortical and sub-cortical regions.

FURTHER CONSIDERATIONS: IS PLANNING JUST WORKING MEMORY FOR THE FUTURE? While correlational approaches to the analysis of imaging data, such as those described above, may provide more task-specific information than standard cognitive subtraction designs, they rarely allow regional CBF changes to be related, unequivocally, to one aspect of task performance. For example, in the

146

OWEN

study by Dagher et al. (1999), described above, while planning undoubtedly increased with complexity of the TOL problems, so did the load on working memory. The possibility remains, therefore, that what was being imaged was not an increase in planning load per se but rather an increase in working memory demands associated with more complex TOL problems. The relationship between planning and working memory has been investigated recently in a large-scale behavioural study in normal control volunteers (Robbins, James, Owen, Sahakian, McInnes, & Rabbitt, 1998). In all, 341 subjects were assessed on the computerized TOL task and on two tests that tap different aspects of spatial working memory. Factor analysis revealed that aspects of performance on all three of these tests loaded, to varying degrees, on a single factor. Thus, measures of both accuracy and (subsequent) thinking time on the TOL test, error score and strategy score on a spatial search test and spatial span score were all closely interrelated. This finding clearly suggests that both short-term spatial memory, as measured by spatial span and “strategic” or “organizational” aspects of spatial working memory are important determinants of normal performance on the TOL planning task. It seems likely that this functional relationship between cognitive planning and aspects of working memory performance reflects some degree of commonality in the neural circuitry that mediates these different processes. Like cognitive planning, in recent years, considerable evidence has accumulated to suggest that many aspects of working memory involve the lateral surface of the frontal lobe. This evidence comes from the study of patients with excisions of frontal cortex (Owen et al., 1990; Owen et al., 1995a; Owen, Morris, Sahakian, Polkey, & Robbins, 1996c; Petrides & Milner, 1982; for review see Petrides, 1989), from lesion and electrophysiological recording work in nonhuman primates (see Goldman-Rakic, 1987; Petrides, 1994, for reviews), and more recently, from functional neuroimaging studies in humans (e.g., Courtney, Ungerlieder, Keil, & Haxby, 1996; Gold, Berman, Randolph, Goldberg, & Weinberger, 1996; Goldberg, Berman, Randolph, Gold, & Weinberger, 1996; Jonides, Smith, Koeppe, Awh, Minoshima, & Mintun, 1993; McCarthy et al., 1994; Owen et al., 1996a; Owen, Evans and Petrides, 1996b; Petrides, Alivisatos, Evans, & Meyer, 1993a; 1993b; Smith, Jonides, Koeppe, Awh, Schumacher, & Minoshima, 1995; Smith, Jonides & Koeppe, 1996; Sweeney et al., 1996; for reviews see Owen, 1997, 2000). Rowe et al. (2001) have recently argued that the key element which distinguishes planning tests such as the TOL from other tasks that depend more heavily on working memory is the emphasis on specific goals (for further discussion, see Morris, Miotto, Feigenbaum, Bullock, & Polkey, 1997; also Morris et al., chapter 8, this volume). In essence, it was suggested, that these tasks only differ from the planning tasks in that they do not require “the construction and evaluation of a path from A to B”. Accordingly, control tasks were developed that included many of the cognitive processes


147

commonly implicated in planning tasks, such as the generation, selection and memory for moves, but did not require a goal. Thus, a “classic” TOL condition, in which volunteers had to plan the best solution to problems requiring a minimum of four moves, was compared to a control condition which required volunteers to generate four moves without being constrained by a goal. Performance of the no-goal control task was associated with activation of exactly the same network of areas, including the dorsolateral frontal cortex, as the TOL task and a direct comparison between the two tasks revealed no residual activity in the prefrontal cortex. Moreover, analogous “imagine, but not move” conditions revealed that this commonality remained irrespective of whether volunteers actually executed moves in either of the two conditions. One interpretation of this finding, favoured by the authors, is that the activity observed in the dorsolateral prefrontal cortex during the TOL task can be accounted for by the components of generating, selecting and remembering moves. An alternative hypothesis, which cannot be disregarded based on those data, is that despite the lack of a task-specific goal (e.g., make a series of moves to reach this particular solution), the non-specific goals of the no-goal control task (e.g., go from this position to another position by generating a series of four moves) were sufficient to “subtract out” any goal-related prefrontal activity. This criticism notwithstanding, it does remain unclear whether the existence of a goal, or indeed any specific cognitive operation, can clearly distinguish the TOL task from tests of working memory at the neural level. Thus, correlational analyses of the interrelationships between various experimental neuropsychological tests have demonstrated that there is a significant association between aspects of working memory and accuracy of planning on the TOL task. Neurosurgical patients with localized excisions of the frontal cortex are impaired on the TOL task and are also significantly impaired at tests that emphasize aspects of working memory. Finally, both planning, assessed using TOL task, and working memory, assessed using a wide variety of different tasks, have been shown to activate a similar region within the mid-dorsolateral frontal cortex.

CONCLUSIONS In recent years, increasingly sophisticated methods of behavioural and physiological assessment, combined with a better understanding of functional neuroanatomy, have led to considerable advances in the conceptualization of cognitive planning, as assessed by the TOL test. For example, functional neuroimaging studies have identified the mid-dorsolateral frontal cortex as an important contributor to planning behaviour, although a specific role in the TOL task has yet to be fully specified. New methods of behavioural testing, such as semi-automatic analyses of eye-tracking

148

OWEN

behaviour are also providing new insights into the components that make up performance on the TOL task. For example, the fact that efficient planning involves selectively biasing gaze towards problem-critical balls suggests that such a strategy may be a critical component in generating viable subgoals as well as in verifying the accuracy of the solution. What emerges, however, is a sense that at both the cognitive and neural level the relationship between planning and working memory has yet to be satisfactorily clarified. One possibility is that the TOL task, having such a strong working memory component, is not in itself a very specific test of planning ability. On the other hand, clear evidence from another test or combination of tests that planning and working memory processes can be functionally dissociated – in patient studies, for example – is lacking. Future studies will clearly need to resolve this issue, perhaps by combining the anatomical resolution afforded by contemporary neuroimaging techniques such as event-related functional magnetic resonance imaging (fMRI) with the “psychological resolution” provided by emerging methods of behavioural measurement, such as eye tracking.

ACKNOWLEDGEMENTS I would like to thank three of my collaborators, Dr A. Dagher, Dr T. L. Hodgson and Dr J. B. Rowe, whose data is described in detail in this chapter.

REFERENCES Adey, W. R., & Meyer, M. (1952). An experimental study of hippocampal afferent pathways from prefrontal and cingulate areas in the monkey. Journal of Anatomy, 86, 58–75. Anderson, J. R. (1993). Rules of mind. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Baker, S. C., Rogers, R. D., Owen A. M., Frith, C. D., Dolan, R. J., Frackowiak, R. S. J., & Robbins T. W. (1996). Neural systems engaged in planning: A PET study of the Tower of London task. Neuropsychologia, 34 (6), 515–526. Barbas, H., & Pandya, D. N. (1989). Architecture and intrinsic connections of prefrontal cortex in rhesus monkey. Journal of Comparative Neurology, 286, 353–375. Bianchi, L. (1922). The mechanism, of the brain and the function of the frontal lobes. Edinburgh: Livingstone. Brodmann, K. (1908). Beitraege zur histologischen Lokalisation der Grosshirnrinde. VI Mitteilung. Die Cortexgliederung des Menscen. Journal of Psychology Neurology, 10, 231–246. Brandt., S. A., & Stark, L. W. (1997). Spontaneous eye movements during visual imagery reflect the content of the visual scene. Journal of Cognitive Neuroscience, 9, 27–38. Clark, A. (1997). Being there: putting brain, body and world together again. Cambridge: MIT Press. Cools, R., Stefanova, E., Barker, R. A., Robbins, T. W., & Owen, A. M. (2002). Dopaminergic modulation of high-level cognition in Parkinson’s disease: the role of the prefrontal cortex revealed by PET. Brain, 125, 584–594. Courtney, S. M., Ungerleider, L. G., Keil, K., & Haxby, J. V. (1996). Object and spatial visual working memory activate separate neural systems in human cortex. Cerebral Cortex, 6, 39–49.


149

Dagher, A., Owen, A. M., & Brooks, D. J. (1999). Mapping the network for planning: a correlational PET activational study with the Tower of London task. Brain, 122, 1973–1987. Dagher, A., Owen, A. M., Boecker, H., & Brooks, D. J. (2001). The role of the striatum and hippocampus in motor planning: A PET activation study in Parkinson’s disease. Brain, 124, 1020–1032. Dehaene, S., & Changeux, J. P. (1997). A hierarchical neuronal network for planning behaviour. Proceedings of the National Academy of Science USA, 94, 13293–13298. Gold, J. M., Berman, K. F., Randolph, C., Goldberg, T. E., & Weinberger, D. R. (1996). PET validation of a novel prefrontal task: Delayed response alternation. Neuropsychology, 10, 3–10. Goldberg, T. E., Berman, K. F., Randolph, C., Gold, J. M., & Weinberger, D. R. (1996). Isolating the mnemonic component in spatial delayed response: A controlled PET 15O-labelled water regional cerebral blood flow study in normal humans. Neuroimage, 3, 69–78. Goldman-Rakic, P. S. (1987). Circuitry of primate prefrontal cortex and the regulation of behavior by representational memory. In F. Plum, & V. Mountcastle (Eds.), Handbook of Physiology, Sec 1, The Nervous System (Vol. 5, pp. 373–417). Bethesda, MD: American Physiological Society. Goldman-Rakic, P. S., Selemon, L. D., & Schwartz, M. L. (1984). Dual pathways connecting the dorsolateral prefrontal cortex with the hippocampal formation and parahippocampal cortex in the rhesus monkey. Neuroscience, 12, 719–743. Harlow, J. M. (1868). Recovery from the passage of an iron bar through the head. Boston Medical Surgery Journal, 2, 327–346. Hayhoe, M. H., Bensinger, D. G., & Ballard, D. H. (1997). Task constraints in visual working memory. Vision Research, 38 (1), 125–137. Hodgson, T. L., Bajwa, A., Owen, A. M., & Kennard, C. (2000). The strategic control of gaze direction in the Tower of London task. Journal of Cognitive Neuroscience, 12 (5), 894–907. Hodgson, T. L., Tiesman, B., Owen, A. M., & Kennard, C. (2002). Abnormal gaze strategies during problem solving in Parkinson’s disease. Neuropsychologia, 40, 411–422. Jonides, J., Smith, E. E., Koeppe, R. A., Awh, E., Minoshima, S., & Mintun, M. A. (1993). Spatial working memory in humans as revealed by PET. Nature, 363, 623–625. Jouandet, M., Gazzaniga, M. S. (1979). The frontal lobes. In M. S. Gazzaniga (Ed.), Handbook of Behavioural Neurobiology (Vol. 2). New York: Plenum Press. Land., M. F., & Furneux, S. (1997). The knowledge base of the oculomotor system. Philosophical Transactions of the Royal Society of London, B352, 1231–1239. McCarthy, G., Blamire, A. M., Puce, A., Nobre, A. C., Bloch, G., Hyder, F., Goldman-Rakic, P. S., & Shulman, R. G. (1994). Functional magnetic resonance imaging of human prefrontal cortex activation during a spatial working memory task. Proceedings of the National Academy of Science, 91, 8690–8694. Middleton, F. A., & Strick, P. L. (1994). Anatomical evidence for cerebellar and basal ganglia involvement in higher cognitive function. Science, 266, 458–461. Middleton, F. A., & Strick, P. L. (1995). Anatomical basis for basal ganglia involvement in working memory. Society for Neuroscience Abstracts, 272, 1. Morris, R. G., Downes, J. J., Evenden, J. L., Sahakian, B. J., Heald, A., & Robbins, T. W. (1988). Planning and spatial working memory in Parkinson’s disease. Journal of Neurology, Neurosurgery & Psychiatry, 51, 757–766. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, B. K. (1993). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367– 1378. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., & Polkey, C. E. (1997). The effect of goal–subgoal conflict on planning ability after frontal- and temporal-lobe lesions in humans. Neuropsychologia, 35, 1147–1157.

150

OWEN

Nauta, W. J. H. (1964). Some efferent connections of the prefrontal cortex in the monkey. In J.M. Warren & K. Akert (Eds.), The frontal granular cortex and behavior (pp. 397–409). New York: McGraw-Hill. Newell, A. (1990). Unified theories of cognition. Cambridge: Cambridge University Press. Newell, A. & Simon, H. A. (1972). Human problem solving. Engelwood Cliffs, NJ: Prentice-Hall. Owen, A. M. (1997). Functional organization of working memory processes within human lateral frontal cortex: The contribution of functional neuroimaging. European Journal of Neuroscience, 9, 1329–1339. Owen, A. M. (2000). The role of the lateral frontal cortex in mnemonic processing: The contribution of functional neuroimaging. Experimental Brain Research, 133, 330–343. Owen, A. M., Downes, J. D., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1990). Planning and spatial working memory following frontal lobe lesions in man. Neuropsychologia, 28, 1021–1034. Owen A. M., James, M., Leigh, P. N., Summers, B. A., Quinn, N. P., Marsden, C. D., & Robbins T. W. (1992). Fronto-striatal cognitive deficits at different stages of Parkinson’s disease. Brain, 115, 1727–1751. Owen, A. M., Sahakian, B. J., Hodges, J. R., Summers, B. A., Polkey, C. E., & Robbins, T. W. (1995a). Dopamine-dependent fronto-striatal planning deficits in early Parkinson’s disease. Neuropsychology, 9, 126–140. Owen, A. M., Sahakian, B. J., Semple, J., Polkey, C. E., & Robbins, T. W. (1995b). Visuo-spatial short term recognition memory and learning after temporal lobe excisions, frontal lobe excisions or amygdalo-hippocampectomy in man. Neuropsychologia, 33, 1–24. Owen A. M., Doyon, J., Petrides, M., & Evans, A. C. (1996a). Planning and spatial working memory examined with positron emission tomography (PET). European Journal of Neuroscience, 8, 353–364. Owen A. M., Evans, A. C., & Petrides, M. (1996b). Evidence for a two-stage model of spatial working memory processing within the lateral frontal cortex: A positron emission tomography study. Cerebral Cortex, 6, 31–38. Owen, A. M., Morris, R. G., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1996c). Double dissociations of memory and executive functions in working memory tasks following frontal lobe excisions, temporal lobe excisions or amygdalo-hippocampectomy in man. Brain, 119, 1597–1615. Owen, A. M., Doyon, J., Dagher, A., & Evans, A. C. (1998). Abnormal basal-ganglia outflow in Parkinson’s disease identified with positron emission tomography: Implications for higher cortical functions. Brain, 121, 5. Pandya, D. N., & Barnes, C. L. (1987). Architecture and connections of the frontal lobe. In E. Perecman (Ed.), The frontal lobes revisited. New York: IRBN Press. Petrides, M. (1989). Frontal lobes and memory. In F. Boller & J. Grafman (Eds.), Handbook of Neuropsychology (Vol. 9, pp. 59–82). Amsterdam: Elsevier. Petrides, M. (1994). Frontal lobes and working memory: Evidence from investigations of the effects of cortical excisions in nonhuman primates. In F. Boller & J. Grafman (Eds.), Handbook of Neuropsychology (Vol. 9, pp. 59–82). Amsterdam: Elsevier. Petrides, M., & Milner, B. (1982). Deficits on subject-ordered tasks after frontal- and temporallobe lesions in man. Neuropsychologia, 20, 249–262. Petrides, M., & Pandya, D. N. (1994). Comparative architectonic analysis of the human and the macaque frontal cortex. In F. Boller & J. Grafman (Eds.), Handbook of neuropsychology (Vol. 9, pp. 17–58). Amsterdam: Elsevier. Petrides, M., Alivisatos, B., Evans, A. C., & Meyer, E. (1993a). Dissociation of human middorsolateral from posterior dorsolateral frontal cortex in memory processing. Proceedings of the National Academy of Science, 90, 873–877.


151

Petrides, M., Alivisatos, B., Evans, A. C., & Meyer, E. (1993b). Functional activation of the human frontal cortex during the performance of verbal working memory tasks. Proceedings of the National Academy of Science, 90, 878–882. Rezai, K., Andreasen, N. C., Allinger, R., Cohen, G., Swayze, V., & O’Leary, D. S. (1993). The neuropsychology of the prefrontal cortex. Archives of Neurology, 50, 636–642. Robbins, T. W., James, M., Owen, A. M., Sahakian, B. J., McInnes, L., & Rabbitt, P. (1998). A study of performance on tests from the CANTAB battery sensitive to frontal lobe dysfunction in a large sample of normal volunteers: Implications for theories of executive function and cognitive ageing. Journal of the International Neuropsychological Society, 4, 474–490. Rowe, J. B., Owen, A. M., Johnsrude, I. S., & Passingham, R. E. (2001). Imaging the components of a planning task. Neuropsychologia, 39, 315–327. Shallice, T. (1982). Specific impairments of planning. Philosophical Transactions of the Royal Society of London, B298, 199–209. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Smith, E. E., Jonides, J. J., Koeppe, R. A., Awh, E., Schumacher, E. H., & Minoshima, S. (1995). Spatial versus object working memory: PET investigations. Journal of Cognitive Neuroscience, 7, (3), 337–356. Smith, E. E., Jonides, J. J., Koeppe, R. A. (1996). Dissociating verbal and spatial working memory using PET. Cerebral Cortex, 6, 11–20. Sweeney, J. A., Minutun, M. A., Kwee, S., Wiseman, M. B., Brown, D. L., Rosenberg, D. R., & Carl, J. R. (1996). Positron emission tomography study of voluntary saccadic eye movements and spatial working memory. Journal of Neurophysiology, 75, 454–468. Watanabe-Sawaguchi, K., Kubota, K., & Arikuni, T. (1991). Cytoarchitecture and intrafrontal connections of the frontal cortex of the brain of the hamadryas baboon (papio hamadryas). Journal of Comparative Neurology, 311, 108–133.

CHAPTER EIGHT

Planning in patients with focal brain damage: From simple to complex task performance Robin Morris Neuropsychology Unit, Institute of Psychiatry, London

Maria Kotitsa Department of Psychology, Institute of Psychiatry, London

Jessica Bramham Department of Psychology, Institute of Psychiatry, London

INTRODUCTION Frontal lobe involvement in the planning and organization of activity has been well recognized over time, but is not necessarily well understood. Early observations of patients with neurological abnormalities of the prefrontal cortex have set the framework for investigating this issue. Such patients have clear problems in regulating everyday behaviour in an orderly fashion, often with devastating consequences (Eslinger & Damasio, 1985; Goldstein et al., 1993; Luria, 1966). The frontal lobes are a heterogeneous region of the brain, and can more formally be divided into primary (motor), unimodal (including the premotor and Brocas area) and heteromodal (prefrontal) cortex. The latter is the region implicated in the control mechanisms of cognition and behaviour, within which there are further subdivisions: the dorsolateral cortex, thought to support reasoning ability, problem solving and executive functioning; the orbitofrontal cortex, involved in social function and emotional processing; the medial frontal cortex which has been associated with affect, emotional processing and motivation (Morris, Kotitsa, Brooks, Rose, Bullock, & Polkey, 2002a). The dorsolateral prefrontal cortex is the region perhaps most clearly involved in planning ability. This is indicated by neuroimaging studies of brain activation during various executive tasks, for example, the Tower of 153

154

MORRIS, KOTITSA, BRAMHAM

London (TOL) (see Owen, this volume, chapter 7). There is also evidence for involvement in subsidiary processes related to organization, such as the attentional and processing aspects of working memory. This research method delineates brain areas that may support executive functioning in general, but group studies of patients with focal brain lesions may help to dissociate specific functions and show how these relate to everyday behaviour. This in turn informs clinical practice regarding expected impairments and their measurement. This is the theme of this chapter, which reviews a series of experiments conducted by the authors and colleagues, using relatively large samples of patients who have undergone neurosurgical treatment resulting in focal lesions within the prefrontal cortex. The experiments range from exploring the components of planning and organizational ability on simple laboratory tasks, such versions of the Tower of Hanoi (TOH), to more complex “ecologically” valid procedures, also using virtual reality to simulate tasks that are akin to those in the real world. To set these studies in context, some illustrative case examples are provided, showing the types of cognitive and behavioural disturbance that exists with frontal lobe lesions. Such cases provide a rich source of information from which to develop notions of the core components of planning ability and how this complex function can fail with prefrontal cortical brain damage.

ILLUSTRATIVE CASE EXAMPLES Single case studies have shown that planning and organizational impairment can occur in the absence of substantial changes in intellectual function or other aspects of cognition. The most classic case is perhaps Phineas Gage, a mining engineer, who sustained a brain injury through a freak accident in which a tamping iron was projected upwards through the front of his skull (Harlow, 1848; 1868). From being a placid and organized worker, his behaviour and personality changed dramatically, such that he was described as “fitful and irreverent”. He ended up drifting through life socially and occupationally incapacitated (Harlow, 1868). Another archetypal case is IR, the sister the of the famous neurosurgeon Wilder Penfield, who underwent neurosurgery to remove a right frontal lobe tumour. This patient retained her intellectual capacity as indicated by the following excerpt from a letter she wrote to Penfield about a year after the operation: It has been a wonderful year with a new life, new strength, new hope . . . I need firmly to convince myself that I am not really slow, nor stupid, not incapable . . . and the rest will come. (Penfield & Evans, 1935, pp. 115–133)

8. PLANNING IN PATIENTS WITH FOCAL BRAIN DAMAGE

155

This crisp intelligent prose contrasts with an anecdote about her organizational ability when Penfield visited her for a dinner party three months later: One day about fifteen months after the operation she had planned to get a simple supper for one guest (W.P.) and four members of her own family. She looked forward to it with pleasure and had the whole day for preparation . . . When the appointed hour arrived she was in the kitchen, the food was all there, one or two things were on the stove, but the salad was not ready, the meat had not been started and she was distressed and confused by her long continued effort alone. (Penfield & Evans, pp. 115–133)

A more recent example is the patient EVR, studied by Eslinger and Damasio (1985), a college graduate and chief accountant of a firm, who had become a comptroller by the age of 32 years. He had neurosurgery to remove a large meningioma in his frontal lobes, the surgical lesion covering a large portion of the orbitofrontal cortex and impinging also on the dorsolateral prefrontal cortex. Despite retaining his high intellectual capacity and an excellent memory, his professional life deteriorated rapidly. He went into partnership with a disreputable former co-worker and his new business failed, rendering him bankrupt. He then drifted through several jobs, his employers finding him “tardy and disorganised” and sacking him in each case. Eslinger and Damasio provide the following description of his behaviour: He needed about two hours to get ready for work in the morning, and some days were consumed entirely by shaving and hair washing. Deciding where to dine might take hours, as he discussed each restaurant’s seating plan, particulars of menu, atmosphere, and management. He would drive to each restaurant to see how busy it was, but even then he could not finally decide which to choose. Purchasing small items required in-depth consideration of brands, prices, and the best method of purchase. (Eslinger and Damasio, 1985, p. 1732)

These examples provide a powerful insight into frontal lobe functions, the central aspect being the ability to create a coherent structure for sequencing activity and its systematic application. This requires a complex set of mental capacities, including the development of strategies, the ability to deal with novelty, adherence to sets of tasks demands, and allocation of resources and “effort” (Shallice & Burgess, 1996). Laboratory-based tasks have made inroads into measuring these capacities in patients with frontal lobe damage, as indicated in the sections below.

156


DEVELOPMENT OF STRATEGIES Development of strategies when confronted with a new task is an integral part of planning or problem solving, but comparatively little is known about the neural basis. This is partly because there have been few studies of strategy formation in patients with brain damage. Also this is an area where functional neuroimaging may not be so successful as a tool for investigation, since the formation of strategies may be a subtle neuronal event and not easily linked to specific temporal neuronal processing. Early evidence for the involvement of the prefrontal cortex comes from a study of eye movements by Karpov, Luria, & Yarbuss (1968). They required participants with frontal brain damage to search through pictures in response to specific commands (e.g., “Is the family rich or poor?”). Normally this generates specific search paths, because the participants plan their sequence of eye movements to search efficiently. However, the patients made more random search paths in comparison to control participants, reflecting their lack of systematic search strategy. Strategy formation has also been explored in patients with focal frontal lesions using a spatial working memory task developed by Morris, Downes, Sahakian, Evenden, Heald, and Robbins (1988). This task consisted of an array of squares presented on a touch-sensitive computer screen. The participant has to search around the array by touching the squares until a blue chip appears from behind the relevant square. Subsequently, the chip moves to a different location and the participant has to search for it again, and so on until all the locations have been targets. A systematic strategy is to search the squares in a particular order, skipping the ones successful in previous searches. This particular strategy has been shown to be deficient in patients with frontal lobe neurosurgical lesions (Owen, Downes, Sahakian, Polkey, & Robbins, 1990; Owen, Morris, Sahakian, Polkey, & Robbins, 1996). This task has been modified into a computer golfing game by transforming the squares into “golf holes”, and the participant is required to predict the hole into which the golfer will drive the ball by touching different holes until successful (see Figure 8.1). After finding the correct hole, the participant then has to search again for another hole “selected” by the golfer for the next drive. The golf game essentially creates the need for a three-dimensional array and transforms the task into a more ecologically valid procedure. Miotto, Bullock, Polkey, & Morris (1996) tested a sample of patients with unilateral prefrontal lesions and found that only a group with right hemisphere lesions had impaired performance. In addition, the task can be used to measure spatial memory performance, since the participant is required to remember the various holes that have been played by the golfer. The right frontal group showed a larger impairment in this regard than the left unilateral frontal group, but this difference disappeared when strategy formation was covaried

Figure 8.1 Set of searches during the Executive Golf task. The top set of panels shows an ordered set of searches for a normal control participant. The second panel shows the more chaotic search of a patient with neurosurgical frontal lobe damage. The black holes signify the hole the participant selects to start the search. The number of different black holes used in a series of searches provides an index of strategy formation.

158


in the analysis of their deficits. This result indicates that impaired strategy formation was contributing significantly to difficulties performing this task, but this was only associated with the right hemisphere lesions. Furthermore, the evidence from Miotto et al.’s (1996) study suggests that the non-dominant hemisphere supports strategy formation. It is therefore possible that this ability is modality specific in relation to hemispheric specialization. This suggestion is supported by the finding that other spatial tasks that require a degree of strategy, such as the Design Fluency (Jones-Gotman & Milner, 1977) and the nonverbal version of the self-ordered task (Petrides & Milner, 1982) are also sensitive to right frontal lobe damage. The converse of verbal strategy formation being associated specifically with the dominant hemisphere has been partly supported by the finding that verbal versions of the above tasks are sensitive to left frontal lesions.

PROBLEM SOLVING ON THE TOWER OF HANOI (TOH) The TOH and its variations, including the TOL, have been convenient tools for investigating problem solving in patients with neuropsychological impairment (Figure 8.2; see also chapter 1, Ward and Morris, Figures 1.1, 1.2, 1.3). The TOH is widely used as a neuropsychological test mainly because it generates tractable problems, and these can be systematically varied in difficulty to suit groups of patients with different ability levels (Goel & Grafman, 1995; Morris et al., 1988; Owen et al., 1990; Shallice, 1982). The TOH also has the advantage of being “well defined” in that it has specified

Figure 8.2 Examples of TOH problems giving the optimal solution to each problem. (a) The solution involves a goal–subgoal conflict, with the first move (small disc to the left) moved away from the direction of the goal state. (b) The solution does not involve a goal–subgoal conflict, with the first move in the direction of the goal state. In the figure (S) represents a small disc; (M) a medium sized disc; and (LL) a large disc.


159

starting and goal stages, with sets of procedures that can transform one state into another (Greeno, 1978). It has an action tree in which the possible states or permutations of arrangements can all be represented, together with moves between them (see Figure 8.2). Problems such as the TOH can more readily be solved by using the heuristic of a means–end analysis, in which the principal strategy is to identify the main differences between the starting and goal position and find ways of reducing these differences. Tasks such as the TOH, lend themselves to being broken down into subgoals, with evidence for this based on the finding that the number of errors decrease in problems as the participant nears the goal state (Egan & Greeno, 1974). This type of approach has been termed “hill climbing” where the movement is towards the overall goal through a series of subgoals. To achieve these subgoals, in turn, different strategies or routine procedures may be employed and these have been explored through experimental study (Simon, 1975).

The goal–subgoal conflict The straightforward strategy of “goal reduction” is generally effective, but it can fail in certain circumstances. One instance is the so-called “goal–subgoal conflict” which is where the most efficient route to a goal state is counterintuitive, in that it initially appears to be going in the opposite direction away from the goal (see also chapter 5 by Ward). This is seen, for example, in the missionary and cannibal problem in which a fictitious set of missionaries and cannibals have to be transported across a river using a boat that can carry only two people (Garnham & Oakhill, 1994). A constraint of the problem is that there must never be more cannibals than missionaries. In this problem, application of a hill-climbing strategy would involve gradually moving people to the far bank in stages. However, a key point in solving this problem occurs when it is necessary to move people back from the far bank. Hence, if distance from the solution is measured by the number of people on the far bank, it is necessary to move away from the solution for several moves in order to solve the problem. Morris, Miotto, Feigenbaum, Bullock, & Polkey (1997a) identified a series of problems on the TOH in which goal–subgoal conflicts occur. An example is shown in Figure 8.2. A three-disc version of the TOH was used, and the resulting “state space” was used to generate the problems. For the TOH, a goal–subgoal conflict involves having first to move a disc away from the final spatial location. In the problem shown in Figure 8.2, the first move is to place the small disc onto the left rod, away from the final goal state position, on the right-hand rod. In Figure 8.2(b), an example is shown where the first move of the small disc to the central peg is “congruent” with the direction of the overall goal. The problems used in this study were presented on a computer

160


fitted with a touch-sensitive screen. The goal state was presented at the top of the screen and the starting arrangement at the bottom. The participants had to move the discs in this bottom array to match the top by touching the disc they want to move and then the desired location. A particular difficulty with goal–subgoal conflict was predicted to occur with frontal lobe damage because of an inability to inhibit the pre-potent response of going in the immediately obvious direction towards the goal. Accordingly, “goal–subgoal conflict” problems were compared to those in which no such conflict existed in patients with left and right focal frontal neurosurgical lesions (Morris et al., 1997a). In addition to normal control subjects, a further group of patients with left and right unilateral temporal lobectomies were included to form a brain damage control group. In this task, accuracy is defined in terms of the number of additional moves beyond the minimum needed to solve each problem (see Figure 8.3). For the patients with frontal damage this yielded a clear increase in moves in the conflict condition, but only for the left frontal group. The effect was seen at the earlier four-move problems, but dissipated when the five-move problems were attempted. For the right unilateral temporal lobectomy patients, there was also a deficit on the goal–subgoal level four problems, but this deficit became more general for the five-move problems. The pattern of deficits thus differed between the two patient groups. Despite finding a deficit in the left frontal group, the diminution of this effect with the later presented five-move problems encouraged Morris et al. (1997a) to consider a different interpretation from reduced response inhibition. One

Figure 8.3 Number of moves above the minimum when comparing congruent and “goal– subgoal” conflict problems at different levels of difficulty.


161

interpretation is in terms of the novelty of being faced with the goal–subgoal conflict. The processing of novelty has been muted as an important function of the prefrontal cortex, for example, involving the creation of new procedures or algorithms to provide the solution (Shallice, 1982; Shallice & Burgess, 1996). In this case, congruent problems may lend themselves to being more easily solved using standard procedures, but conflict problems may require a newly developed set of routines. This explanation accounts for why the initial deficit dissipates, if the frontal lobe patients eventually “catch up” by developing routines that can be used to cope with goal–subgoal conflict. If this explanation is correct and the deficit is not due to response inhibition, then a further prediction would be that patients with frontal damage would not show a pattern of premature responding. In the study, thinking time measures were taken, including the time taken before the first move, i.e., the planning time. In addition, a psychomotor control condition was used which requires the participant to move the discs according to the same sequence as when they were solving the problems but being “led” by the computer. Subtracting these times from the latencies during problem solving provides a purer measure of thinking time. Both measures of thinking time did indeed show that the frontal lobe patients took the same time as the controls to plan their moves in all conditions. For the patients with temporal lobe damage, deficits in problem solving were also apparent in the right unilateral temporal lobectomy group. Here, however, the deficits persisted and became more generalized for the five-move problems. Since this group is known to have impairments in visuo-spatial memory, one interpretation was that the deficit was due to memory problems. To test this, Morris et al. (1997a) used a memory control condition in which the patients were shown a sequence of moves on a single TOH “plinth” and had to repeat these moves from memory. The number of moves to be remembered increased until the patient failed consistently. The right unilateral temporal lobectomy patients were found to be impaired on this task. Furthermore, when the measure was covaried in the TOH accuracy analysis, the deficit on both difficulty level three and five problems disappeared. However, this did not occur for the left frontal group. This strongly suggests that memory impairment was “driving” impairment in patients with temporal but not frontal lesions.

State action and search selection equivocation An alternative approach to problem solving is to trace down the state action tree, searching for solutions until the goal state is achieved. This approach is made easier with a limited sized state action tree such as in a three-disc TOH. However, computationally it is still intractable if done systematically, so heuristic devices have to be used as well, such as the means–end analysis, as

162


indicated above. Nevertheless, one of the features of problem solving on this task is to follow a sequence of moves until a solution is found or until a point is reached where the sequence is seen to be unsuccessful. Different search paths can then be compared for the efficiency and the optimal sequence selected. The need to trace down different search paths may impose a substantial memory load. This load may increase with the size of the solution path, or the degree to which different solutions have to be compared. In the three-disc version, the maximum possible direct distance between a starting and goal state is seven moves, but this increases with less efficient solutions. This aspect is highlighted when there are two main plausible routes between the starting and goal state (see Figures 8.4 and 8.5). This sets up a need to compare different solutions for efficiency, a demand that has been termed selection equivocation (Morris et al., 1997a). The two main routes are distinct because of the characteristics of the problem space. This can be split into three main domains, each of which is defined by the position of the

Figure 8.4 Example of Tower of Hanoi (TOH) problem involving two major paths of similar length between the start and goal state. In the figure (S) represents a small disc; (M) a medium sized disc; and (LL) a large disc. The total problem space is given, the connections representing moves between the various arrangements.


163

Figure 8.5 Example of TOH problem involving two major paths of dissimilar length between the start and goal state.

larger disc (left, middle and right). The different domains are joined in a ring such that a feature of the problem space is that any solution requiring more than four moves must involve passing between these domains. The two main routes are thus moving either clockwise or anticlockwise around the ring to achieve the goal state. By selecting different positions of starting and goal states it is possible to create problems in which the two main paths are either similar or dissimilar in the length (see Figures 8.4 and 8.5). A prediction is that if a patient has difficulties in tracking down the solution paths, this will produce less efficient solutions where the paths are dissimilar and the patient will be more likely to opt for the longer and less efficient path. This was investigated in patients with unilateral frontal and temporal lesions (Morris, Miotto, Feigenbaum, Bullock, & Polkey, 1997b), the sample as in the experiment by Morris et al. (1997a). Here it was expected that patients with right temporal lobectomy may have special problems with selection equivocation because of their spatial memory deficits. On the other hand, frontal lobe damage may result in problems with selecting between alternative paths, due to executive dysfunction. A direct comparison between similar and dissimilar length problems is shown in Figure 8.6. This shows that for the right temporal lobectomy patients there was less accuracy for the dissimilar length problems as predicted. For the frontal lobe patients, a deficit

164


Figure 8.6 Number of moves above the minimum when comparing TOH problems with main paths between start and goal of similar or dissimilar length.

was seen only for the patients with right lesions, but this applied to both the similar and dissimilar conditions. Hence being hindered by difficulties in tracking down the search paths was not a particular feature of the impairment in these patients. These results suggest that selection equivocation is not a specific problem with frontal lobe damage, but relates more specifically to right temporal lobe damage. Since such patients have spatial memory impairment (see Morris & Parslow, 2004), it seems that an inability to track down the solution paths and decide on the most efficient may depend on remembering the routes taken for the purposes of comparison and later execution.

Conclusions These two studies illustrate the cognitive components that can be considered in planning tasks such as the TOH and also how these can dissociate in patients with focal brain lesions (see Table 8.1). A pattern emerged with the goal–subgoal conflict impairment associated with left frontal lesions, but unrelated to memory. A general deficit with more lengthy problems occurred with right frontal lesions. Finally, the right temporal lobectomy patients showed impairment on the TOH and this appeared to be related to memory difficulties, giving rise to impairments with dissimilar length problems in the selection equivocation experiment. Differences in the pattern of deficits according to whether there were left or right hemisphere frontal lesions, are notable because the same patients


165

TABLE 8.1 Summary of main findings on Tower of Hanoi (TOH) with patients with frontal lobe lesions. The left frontal lobe patients show impairments with goal–subgoal conflicts. The right frontal lobe patients show impairment when the number of moves need to solve a problem increases to a high level (high complexity). The right temporal lobectomy patients are impaired with selection equivocation (which increases memory load), or high complexity, suggesting their impairment is due to spatial memory difficulties Types of impairment Goal–subgoal conflict

Selection equivocation

High complexity

Memory span

Yes* No No No

No No No Yes

No Yes No Yes

Yes No No Yes

Left frontal Right frontal Left temporal Right temporal * Unrelated to memory span.

were tested in both the Morris et al. (1997a, 1997b) studies. The notion of the left prefrontal cortex being involved in responding to novelty has been introduced by Shallice (1988). This would explain impairment associated with problems that do not require more routine or obvious procedures, such as using a simple hill-climbing strategy, as in the congruent problems used by Morris et al. (1997a). On the other hand, the longer problems used in the selection equivocation experiment revealing the impairment in the right frontal patients may require a different explanation. One possibility is that the longer problems invoke a range of predominately spatial strategies and that these are supported in turn by the right prefrontal cortex. Some support for this comes from the finding of spatial strategy formation impairment in an overlapping patient sample of right frontal patients (Miotto et al., 1996). Also there is evidence for poor organizational strategies or self-monitoring on visuo-spatial tasks in right frontal patients, as shown by Petrides and Milner (1982) on the self-ordered pointing task. A number of TOH approaches have been identified, such as moving the largest disc into position first, followed by the next largest disc. In all, four main strategies have been identified, including goal recursion, a move pattern strategy, and two perceptual strategies, as outlined by Simon (1975). Clearly further studies are needed to determine the development of these strategies in patients with different types of focal neurological damage and whether they fail to occur efficiently in patients with frontal lobe lesions.

166


PLANNING AND ORGANIZATIONAL ABILITIES INVESTIGATED THROUGH SIMULATION A criticism of the approach described above is that the mental activity associated with solving TOH problems is somewhat removed from planning “in the real world”. In order to circumvent this problem, there have been several attempts to develop tasks incorporating activities that a person might perform in everyday circumstances. For example, Goel, Grabman, Tajik, Gana, and Danto (1997) used a financial planning task in a single case study of a patient with frontal lobe damage and Phillips and colleagues (see chapter 6, this volume) have developed a “party planning” task to explore the functioning of older adults. In a similar vein, Shallice and Burgess (1981; see also, Goldstein et al., 1993) designed the Multiple Errands Task (MET) in which they required their participants to carry out a number of errands in Lambs Conduit Street, close to the National Hospital in London. For example, they had to buy certain items such as a brown loaf and a packet of fruit pastilles. Other tasks were more complex, such as obtaining the information needed to send a postcard. Particular rules were built into the whole task, such as having to spend as little money as possible. This test exposed planning and organizational deficits in otherwise high-functioning patients with frontal lobe lesions. Another, called the Six Elements Task (SET), required patients to organize themselves to carry out six tasks in a specific time limit, without breaking certain rules. This test also proved sensitive to deficits in patients with frontal lobe damage, who nevertheless performed normally on conventional tests of executive functioning, such as the Wisconsin Card Sorting test. These experimental studies have also given rise to standardized tests such the Zoo Map test and the Modified Six Elements test, which are now in routine clinical use as part of the Behavioural Assessment of Dysexecutive Syndrome (Wilson, Alderman, Burgess, Elmsie, & Evans, 1996).

Virtual Planning Test (VPT) To explore planning ability, a task in the form of a boardgame was constructed by Morris and Miotto (1998). This involves the participant planning activities for four days (Monday to Thursday), which all had to be carried out before a fictional trip abroad (starting on the Friday). The participant is provided with a set of 28 activity cards that are laid before them. They have to select four of these cards for each day and do so by transferring them on to an “execution” board. Once the activities have been placed on this board, they are not allowed to change their minds and must go on and select activities for the next day, which necessarily involves planning. The participant is given a set of instructions outlining certain tasks that have to be completed within


167

the four days, based on the set of activity cards. The memory load is reduced for this task by providing a summary card (see Figure 8.7). Built into this task are various components to test different aspects of planning which are relevant to dysfunction in patients with frontal lobe damage: (1) Complexity. Each activity card contains a specific activity. These can also be grouped together into a sequence to produce a set of activities needed to complete an overall task (see Figure 8.8). For example, posting a letter to a friend would require two activities: buying an envelope, and going to the post office to post the letter. The complexity of each task can be manipulated by varying the number of preparatory activities for the final activity, either using one or two items. This enabled the study to explore the degree to which complexity either ameliorates or exacerbates the degree of impairment associated with frontal lobe damage. (2) Context. The activities are divided into two types: those that just have to be done in the time before the trip, and others specifically pertaining to the fictional trip abroad. Hence they were either related to the current context (e.g., “pay your electricity bill by Friday”), or a removed scenario (e.g., “buy a suitcase you will need for the trip”). (3) Time specificity. This is related to the timing of activities. Some of these have to be done at a particular time, on a specified day (e.g., “Go to the bank to get some travellers’ cheques for your trip. This has to be done on Thursday morning”). Others have to be completed, but there is no time constraint other than by the end of the four days (e.g., “Pay your telephone bill by Friday”). The grouping of individual activities into sequences and also the time specificity of others require the participants to plan ahead in order to be able to complete all the tasks. In addition, distracter activities are included to see whether the participants would be “lured” away from the given tasks. Some of these were associated with a trip abroad and some with the current context (e.g., “stay indoors and take it easy for a while”, or “buy a film for your camera”). This test was administered to a mixed sample of nine left frontal, ten right frontal and six bifrontal patients. As one group, the frontal patients showed a clear impairment overall in terms of the number of times they failed to do an activity, either at or by the appropriate time. For the complexity variable, there was no impairment when the units had two preparatory activities, but there was impairment with only one act. In contrast, there was no difference in terms of level of impairment when considering the trip-related or current context. Nevertheless, this distinction did yield differences for the distracter items. Trip-related distracter did not differentiate the patients from the

168


Figure 8.7 Activities for completion as listed on the summary card used in the Virtual Planning Test (VPT).

Figure 8.8 Examples of activity units of different complexities. The top example has one preparatory act, and the bottom illustration has two preparatory acts.

controls, but there was a difference with the same-context items. This might suggest that the patients were more easily distracted when the competing activities fitted with their immediate context. Finally, there was no differential deficit according to whether the activities had to be performed at a particular time or at any time. This might not have been predicted, given that other studies suggested that prospective memory impairments are associated with frontal lobe damage (Cockburn, 1995). However, this result may reflect the fact that the time specificity of the action


169

was always provided by the activity cards (e.g., “Collect your theatre tickets from the box office on Wednesday morning”). Here, the participant is cued to produce the action at the appropriate time, rather than having to generate their response from internally laid down markers.

Theoretical implications In conclusion, this task shows significant planning and organizational deficits in patients with frontal lobe lesions. Further analysis showed that this was not dependent on the side of lesion. The main finding of impairments being associated with smaller units of activity can be considered in relation to the types of factors that cue action. A way of interpreting this finding is that individual activities associated with the larger units are more likely to be carried out because they become more dominant overall. This ameliorates the planning deficit associated with frontal lobe damage. This, however, is at the expense of smaller units, which are essentially neglected. Such dominance-led prioritization can be considered in relation to contention scheduling models of action (Shallice, 1982, 1988), as illustrated in Figure 8.9. A simple contention scheduling mechanism may consist of a number of schemata that represent the basis for action. These are activated via a triggering mechanism, which, in turn, is cued by perceptual input, or internally driven thought processes. Where schemata are linked through units of activity, there is mutual excitation at the level of triggering for each schema. Hence, a larger grouping of activities is more likely to mutually trigger the component activities. This makes a natural propensity for larger units to dominate. Nevertheless, this attribute of contention scheduling may be overridden by executive control. In terms of the Shallice and Norman

Figure 8.9 Contention scheduling: units of activity are mutually excitatory and the likelihood that a particular schema will be activated is increased if it forms part of a unit. Larger units become dominant because of the greater probability of reciprocal activation.

170


Figure 8.10 Contention scheduling: a contextual network is activated (in this case Context 2) and this increases the probability that the triggering mechanism will activate certain schema given perceptual input. The context activation is also cued by perceptual input in the absence of executive control.

model (Shallice, 1982, 1988), this function is served by the supervisory attentional system (SAS), which is able to reset the priority for schemata activation, including suppressing the activation of schemata that are inappropriate. In the case of frontal lobe damage, executive control mechanisms are weakened, such that the contention scheduling mechanism operates in a more “freewheeling” manner, hence the neglect of smaller units of activity. This model explains a number of behavioural features of frontal disorder, including the degree to which patients fail to attend to minor details when performing psychometric tasks, a characteristic feature of the frontal lobe syndrome. Another main finding was that the same-context distracter tasks were more likely to be selected by the frontal lobe patients. This finding is consistent with the manner in which the behaviour of these patients tends to be determined more extensively by proximal cues in the environment, rather than those generated internally. Hence, in a phenomenon such as perseveration, such triggering could occur by the same immediate contextual cues. The activity itself could also keep in place the contextual cues and hence result in a feedback loop in which the cycle repeats. Also there is the degree to which patients are distracted by immediate perceptual input. Again, this type of impairment can be considered in relation to contention scheduling mech-


171

anisms where “contextual proximity” may play a role in altering the thresholds for the components of the triggering mechanism. This is illustrated in Figure 8.10, where the current context (in this case Context 2) is represented diagrammatically as a network of activation. This in turn has associative links with the triggering mechanism to lower the threshold of activation for contextually relevant activities. Hence, when the contention scheduling mechanism is allowed to operate freely, it will tend to result in activities related to the immediate context. The thresholds will change when the activated contextual network alters, but this change will occur either when driven heavily by perceptual input or when modulated by executive control. In the case of frontal lobe damage, the context may remain static, particularly when driven by the same stimulus input, but also direct activities related to that context. Thus the model explains both the difficulties that patients have in shifting context, but also the degree that they tend to be distracted by associated activities, even if these are inappropriate.

VIRTUAL REALITY EXPLORATION OF STRATEGY FORMATION, RULE BREAKS AND PROSPECTIVE MEMORY The Virtual Planning Test (VPT) described above is an attempt to simulate everyday planning, albeit in a somewhat artificial scenario using a “boardgame.” By requiring the participant to think about a series of actions, it does invoke the types of mental processes a person would use in planning ahead. However, the more static nature of a “pencil and paper” task or “boardgame” is to some extent not quite the same as the interactive nature of engaging in a task carried out in the real world. “Real-life” tests such as multiple errands obviate these caveats, but these are difficult to use systematically over large groups of patients and have limitations in terms of experimental control and replicability. More recently, we have been developing virtual reality based procedures which we used to explore various facets of planning and problem solving in patients with focal brain lesions (Morris et al., 2002a; Morris, Kotitsa, Bramham, Brooks, & Rose, 2002b). These tests are presented on desktop computers, with the participant navigating around a virtual environment using a joystick and responding to objects in the environment by touching them on the visual display unit.

The Bungalow task This task consists of a virtual reality house, a bungalow, which has four rooms and a hallway (Figure 8.11). The bungalow is used as a setting for a test of planning and organization, which invokes various cognitive components as follows:

172


Figure 8.11 The four rooms and hallway from the Bungalow task shown in plan view. Views of each room and distributed furniture are also shown. The hallway is shown at the bottom.

(1) The participants are told that they are a “removal person” and have to move around the rooms selecting furniture that has to be taken to a different house. The furniture has to be selected appropriately for the rooms in the new house and must be collected in a certain order, according to their category. So all the furniture for the lounge of the new house must be collected first, followed by the dining room, nursery, kitchen, study, music room, bedroom and hallway. The furniture for these rooms is distributed throughout the bungalow, such that


173

the participant must adopt a certain strategy to select the furniture efficiently and in order. (2) A number of tests of prospective memory are embedded in the task. These include: (a) Activity related: here they have to remember to shut the front door on entering the bungalow and they should always shut the door when they exit a particular room (room two); (b) Time related: the participant is told that a van is coming to the front door to take the selected furniture. The doorbell is not working, so they should visit the front door every five minutes. They are provided with a clock and an updated time of arrival of the van every five minutes; (c) Event related: arranged round the rooms are specific items of furniture that contain glass. When encountered, these should not be selected for removal but should be labelled with a “fragile” notice (Figures 8.12 and 8.13). These instructions are explained to the participants and they are also provided in summary on a cue card that can be referred to throughout the task. Morris et al. (2002b) have administered this task to patients with unilateral prefrontal cortex lesions. They found that the patients take the same amount of time to complete the task, although the frontal group visit fewer rooms, reflecting a less efficient strategy. The main strategy in this task consists of visiting the rooms in a certain order, repetitively throughout. The purpose is to move through all the rooms of the bungalow selecting all the furniture for the first room of the new house and so on until all the rooms have been completed. The most efficient mode is to keep the order of room searches the same, and simply repeat this order for every new category of furniture. Analysis of normal controls suggests that there are two dominant orders, as illustrated in Figure 8.14. One consists of visiting room 1, followed by 2, 3 and 4, whilst the other involves visiting room 1, 4, 3 and 2 in turn. Examples of deviations from these patterns are also shown in Figure 8.14, illustrating the more chaotic nature of the responses of

Figure 8.12 Different views from the Bungalow task relating to prospective memory. (1) A view of the front door from the hallway. This door must be shut on entering the bungalow (event related) and also checked every five minutes (time related. (2) The door of room two, which must be shut on exit (event related).

174


Figure 8.13 Event-related prospective memory: the participant has to “place” fragile notices on items which have glass whenever they are encountered. This illustrations shows a microwave and personal computer being labelled.

patients with frontal lobe damage. A strategy score was computed by analysing the degree to which the route of the participant differed from both strategies and taking the highest strategy score of the two outcome scores. This revealed a significant deficit in strategy formation in patients with both right and left hemisphere lesions. In addition to developing a strategy, the participants had to follow the general rule of collecting the furniture for each new room in turn. A measure of deviation from this general rule was computed. This showed that the frontal lobe patients were much less efficient at following the rule. In particular, the patients with left frontal lesions showed greater impairment. Finally, the prospective memory measures yielded a specific pattern of results. Here, the patients were unimpaired on activity-related measures, including remembering to shut the front door and that of the second room. There were impairments in both right and left frontal patients, however, on the event-related measure, with the patients more likely to forget to put “fragile” notices on the glass items (see Figure 8.13). In addition, for the time-based prospective memory measure, remembering to go to the door every five minutes, there was a significant impairment in both patient


175

Figure 8.14 Strategy formation on the Bungalow task. The left panels show the two main strategies of moving round the rooms to collect the furniture. On the right panels are the paths of two patients with frontal lobe damage, showing their lack of strategy.

groups where they were substantially later than controls in arriving at the door. Such deficits show how this more ecologically valid test is sensitive to the planning impairments of these patients. The deficits on the Bungalow task related to strategy formation, rule breaking and event and time based prospective memory, illustrating the multi-componential pattern of deficits found on more complex tasks that might simulate real-life activities.

The Warehouse task Just as the Bungalow test was designed to mirror the Multiple Errands Task (MET), a further virtual reality test has been developed using the same principles as the Six Elements Task (SET) developed by Shallice and Burgess (1991). The principle is that effort has to be distributed between different elements of a task, following certain constraints in terms of sequencing of activity. To create a virtual reality version that is relevant to everyday functioning,

176


a warehouse scenario was used, forming the basis of the Warehouse Six Elements Test (WSET). This features a factory warehouse room containing a series of shelves storing cylindrical and box-shaped objects (see Figure 8.15). The participants are trained to do three main tasks after selecting an object: putting objects on a trolley; placing objects on a conveyor belt; weighing objects then placing them on a trolley (see Figure 8.15). In order to explore the ability to distribute effort between the three tasks, they are given a time limit of ten minutes and asked to do an even amount of each task. In order to build in a sequencing constraint, they were required either to collect boxes or round items, as part A and B of each task. The constraint was that part B should never follow immediately after part A of a particular task. The same patients were tested as those in the Bungalow task experiment. There were two main measures: the degree to which effort was allocated between the different components and a rule-break score. The allocation of effort was defined as the mean of the unsigned deviation from a ratio of 1/6, i.e., 0.1667 being the perfect score for each task. This evenness ratio showed a significant difference between both right and left unilateral frontal lobectomies and control participants, the evenness ratios being 0.56, 0.44

Figure 8.15 The Warehouse Six Elements Test (WSET). The large panel shows the general layout of the room. Panels 1–4 show the shelves, the trolley, the weighing machine and the conveyor belt.


177

and 0.17 respectively. Also a rule-break measure was computed by taking the number of categories where the main rule was broken. Here, only the right frontal group showed a significantly higher level of rule breaks than the controls. It should be noted that these patients did not necessarily exhibit impairments on conventional psychometric tests. For example, Morris et al. (2002b) also tested the patients on the Zoo Map test (described above) and this did not reveal any significant differences from the control subjects.

CONCLUSIONS The studies reviewed above show how investigations of planning and organization can start with simple laboratory tasks and develop more complex procedures to incorporate the more dynamic aspects of everyday behaviour. Simple procedures show how strategy formation is impaired in patients with frontal lobe damage, and also give some indication of the factors that affect planning and problem solving. More complex tasks show how these can be also investigated in an “ecologically valid” form. The studies reviewed above are the first to use virtual reality to investigate executive dysfunction in brain damaged patients and show how relatively complex tasks can be used to explore more detailed cognitive aspects of planning ability. Virtual reality offers the possibility for experimental control not so easily afforded by using “real world” tasks. Whilst this approach is experimental in nature, in the future it may be converted into standard clinical tools for neuropsychological investigation given that it has been shown to be more sensitive to brain damage than conventional tests. Furthermore the results of such neuropsychological studies reviewed above will continue to inform theory regarding the cognitive underpinning of planning ability.

REFERENCES Cockburn, J. (1995). Task interruption in prospective memory: A frontal lobe function? Cortex, 31, 87–97. Egan, D. E., & Greeno, J. G. (1974). Theory of rule induction: Knowledge acquired in concept learning, serial pattern learning and problem solving. In L.W. Gregg (Ed.), Knowledge and cognition. Potomac, MD: Lawrence Erlbaum Associates, Inc. Eslinger, P. J., & Damasio, A. R. (1985). Severe disturbance of higher cognition after bilateral frontal lobe ablation: Patient EVR. Neurology, 35, 1731–1741. Garnham, A., & Oakhill, J. (1994). Thinking and reasoning. Oxford: Blackwell. Goel, V., & Grafman, J. (1995). Are the frontal lobes implicated in “planning” functions? Interpreting data from the Tower of Hanoi. Neuropsychologia, 33, 623–642. Goel, V., Grafman, J., Tajik, J., Gana, S., & Danto, D. (1997). A study of the performance of patients with frontal lobe lesions in a financial planning task. Brain, 120, 1805–1822.

178


Goldstein, L. H., Bernard, S., Fenwick, P. B. C., Burgess, P. W., & McNeil, J. (1993). Unilateral temporal lobectomy can produce strategy application disorder. Journal of Neurology, Neurosurgery & Psychiatry, 56, 274–276. Greeno, J. G. (1978). Nature of problem solving abilities. In W. K. Estes (Ed.), Handbook of learning and cognitive processes (Vol. 5 pp. 239–270). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Harlow, J. M. (1848). Passage of an iron rod through the head. Boston Medical Surgery Journal, 39, 389–393. Harlow, J. M. (1868). Recovery from the passage of an iron bar through the head. Paper presented at the meeting of the the Massachusetts Medical Society. Jones-Gotman, M., & Milner, B. (1977). Design fluency: The invention of nonsense drawings after focal cortical lesions. Neuropsychologia, 1, 653–674. Karpov, B. A., Luria, A. R., & Yarbuss, A. A. (1968). Disturbances of the structure of active perception in lesions of the posterior and anterior regions of the brain. Neuropsychologia, 6, 157–166. Luria, A. R. (1966). Human brain and psycholological processes. New York: Harper & Row. Miotto, E. C., Bullock, P., Polkey, C. E., & Morris, R. G. (1996). Spatial working memory in patients with frontal lesions. Cortex, 32, 613–630. Morris, R. G., & Miotto, E. C. (1998). Virtual planning in patients with frontal lobe lesions. Cortex, 34, 639–657. Morris, R. G., & Parslow, D. (2004). Neurocognitive components of spatial memory. In G. L. Allen (Ed.), Remembering where (pp. 217–247). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, B. K. (1996). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367–1378. Morris, R. G., Downes, J. J., Sahakian, B. J., Evenden, J. L., Heald, A., & Robbins, T. W. (1988). Planning and spatial working memory in Parkinson’s disease. Journal of Neurology, Neurosurgery & Psychiatry, 51, 757–766. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., & Polkey, C. E. (1997a). The effect of goal–subgoal conflict on planning ability after frontal and temporal lobe lesions in humans. Neuropsychologia, 35 (3), 1147–1157. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., & Polkey, C. E. (1997b). Planning ability after frontal and temporal lobe lesions in humans: The effects of selection equivocation and working memory. Cognitive Neuropsychology, 14 (7), 1007–1027. Morris, R. G., Kotitsa, M., Brooks, B., Rose, D., Bullock, P., & Polkey, C. E. (2002a). Virtual reality investigations of planning ability following focal prefrontal cortical lesions. Paper presented at the British Psychological Society Cognitive Section Annual Conference. September, Canterbury. Morris, R. G., Kotitsa, M., Bramham, J., Brooks, B., & Rose F. E. (2002b). Virtual reality investigation of strategy formation, rule breaking and prospective memory in patients with focal prefrontal neurosurgical lesions. Proceedings of the 4th International Conference on Disability, Virtual Reality and Associated Technologies (pp. 101–108). Vezzprem, Hungary. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1990). Planning and spatial working memory following frontal lesions in man. Neuropsychologia, 28, 1021–1034. Owen, A. M., Morris, R. G., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1996). Double dissociations of memory and executive functions in working memory tasks following frontal lobe excisions, temporal lobe excisions or amygdalo-hippocampectomy in man. Brain, 119, 1597–1615.


179

Penfield, W., & Evans, J. J. (1935). The frontal lobe in man: A clinical study of maximum removals. Brain, 58, 115–133. Petrides, M., & Milner, B. (1982). Deficits on subject ordered tasks after frontal- and temporallobe lesions in man. Neuropsychologia, 20, 249–262. Shallice, T. (1982). Specific impairments of planning. Philosophical Transactions of the Royal Society, London, B298, 199–209. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T., & Burgess, P. W. (1981). Deficits in strategy application following frontal lobe damage in man. Brain, 114, 727–741. Shallice, T., & Burgess, P. W. (1996). The domain of supervisory processes and temporal organisation of behaviour. In A. C. Roberts, T. W. Robbins, & L. Weiskrantz (Eds.), Executive and cognitive functions of the prefrontal cortex. Philosophical Transactions of the Royal Society of London, 351, 405–412. Simon, H. A. (1975). The functional equivalence of problem solving skills. Cognitive Psychology, 7, 268–288. Wilson, B. A., Alderman, N., Burgess, P. W., Emslie, H., & Evans, J. J. (1996). Behavioural assessment of the dysexecutive syndrome. Bury St Edmunds: Thames Valley Test Company.

CHAPTER NINE

Planning and the brain Jordan Grafman Cognitive Neuroscience Section, National Institute of Neurological Disorders and Stroke, Bethesda MA, USA

Lee Spector Department of Cognitive and Computational Sciences, Hampshire College, Amherst MA, USA

Mary Jo Rattermann Department of Psychology, Franklin and Marshall College, Lancaster PA, USA

INTRODUCTION Plans are an ubiquitous part of human activity. A plan can be defined as a structured event series that generally contains one or more goals. Plans range from the short term and motoric, such as planning a sequence of key presses (Pascual-Leone et al., 1993) to the long term and cognitive, such as deciding on the steps required for air traffic controllers to land a specific airplane (Suchman, 1987). How plans are developed and executed has been the focus of study in artificial intelligence (AI) (Allen, Kautz, Pelavin, & Tenenberg, 1991; Hammond, 1994), cognitive science (Friedman & Scholnick, 1997; Hoc, 1988), and neuropsychology (Owen, 1997). In their prescient book on planning, Miller, Galanter, and Pribram (1960) revealed the difficulty that neuropsychology might have with identifying which brain structures would be concerned with planning as defined by contemporary computer science terminology, eventually admitting: “The relation between computers and the brain was a battle the authors fought with one another until the exasperation became unbearable.” The responsibility for this difficulty may partly lie in the different methods used to investigate planning by each discipline. Besides their differences, each discipline’s methods has particular weaknesses. For example, Langley and Drummond have recently decried the non-experimental basis of much of the AI literature on planning (Langley & Drummond, 1990). They have argued for the development of testable hypotheses that can be experimentally addressed such as: “What are the resources required to 181

182

GRAFMAN, SPECTOR, RATTERMANN

generate a plan?” “In reacting to an unexpected event, how much sampling of the environment is done?” “What is the ratio of deliberation to execution (and what does that ratio depend upon)?” “How do subjects modify stored plans versus constructing entirely new plans?” Many AI and cognitive researchers have also noted the similarity between plans and other knowledge structures such as story grammars, themes, action sets, cases, schemas and scripts (Schank & Abelson, 1977). All of these knowledge structures represent sequential and structured information that must be sustained in some active state over time for processing (Kolodner, 1993). In this chapter we are concerned with identifying what is established in the cognitive neuroscience study of planning. We will mainly draw upon results of studies that explicitly tested planning although we will cite data from a few studies that investigated script and story processing. Our goal in this chapter is not to describe the intimate role that objects, scenes, lexical knowledge, and motor actions have in constructing and executing a plan, but to provide a description of the cognitive components of plan-specific knowledge that can be mapped to brain. We suggest that the crucial components of plan-specific knowledge are primarily stored in the prefrontal cortex with plan execution assisted by motor processes stored in the basal ganglia and frontal lobes (Grafman, 1989, 1994, 1995; Grafman & Hendler, 1991). Before reviewing the cognitive neuroscience investigation of planning, we will introduce the cognitive and computational science perspectives on planning setting the groundwork for our claims about which plan-specific components of knowledge are distinctively stored in the human prefrontal cortex.

COGNITIVE AND COMPUTATIONAL PERSPECTIVES ON PLANNING Cognitive scientists generally describe planning as the process of formulating an abstract sequence of operations intended for achieving some goal (HayesRoth & Hayes-Roth, 1979; Scholnick & Friedman, 1987). The representation of this sequence is called a plan (Wilensky, 1983). A plan can be represented internally (in the planner’s mind) or externally (e.g., a blueprint, travel route). There are two predominant views of planning within cognitive psychology: successive refinement models and opportunistic models. Successive refinement models propose that planning is a top-down hierarchical process, much like a computer program, that controls the order in which a series of operations can be performed (Miller et al., 1960; Newell & Simon, 1972; Sacerdoti, 1975). Opportunistic models propose that planning is a data-driven process which can operate concurrently at several different levels of abstraction, with decisions at any level affecting subsequent decisions at both higher and lower levels (Hayes-Roth & Hayes-Roth, 1979).

9. PLANNING AND THE BRAIN

183

The view of planning as successive refinement had its beginning in the work of Miller et al. (1960). They proposed that a plan is “any hierarchical process in the organism that can control the order in which a sequence of operations can be performed” (p. 16). These plans usually include hierarchically organized subplans, which can include further subplans, down to the level of motor action (Das, Kar, & Parrila, 1996). At each level or subplan, the planner executes a TOTE (Test-Operate-Test-Exit) unit, where the planner tests to see if a goal is satisfied. If it is not, she operates to achieve the goal, tests the efficacy of the operation, and then if the goal is met, exits. Upon exiting, the planner moves to the next step in the sequence (Scholnick & Friedman, 1987). This hierarchical, top-down view of planning is evident in many cognitive models, including planning as problem solving within the SOAR (State, Operator, and Result) architecture (Rosenbloom et al., 1993), Schank and Abelson’s view of plans as a general mechanism underlying the formation of scripts (Schank & Abelson, 1977), and in views of planning drawn from artificial intelligence (AI) (Fikes & Nilsson, 1971; Sacerdoti, 1975). An alternative to the hierarchical, successive refinement models of planning are opportunistic models, such as that proposed by Hayes-Roth and Hayes-Roth (1979). They proposed that at each point in the planning process a planner’s current decision affects the opportunities available and the decisions that must be made later in the development of a plan. Thus, plans grow incrementally as each new decision is incorporated into and revises previous decisions, creating a multi-directional, revisionary planning process. Thus, decisions can be made at any level of abstraction at any point in the planning process. In some domains and planning situations this process will begin at a high level of abstraction and the plan will develop in an orderly, top-down expansion of goals and subgoals, much like in planning by successive refinement. In other domains and planning situations, however, this process will begin as a series of concrete local decisions and the planning process will move between highly abstract decisions and concrete local decisions, often without an overall framework for the decision making process (Pea & Hawkins, 1987). This basic model of opportunistic planning became the inspiration for many cognitive psychologists studying planning in adults and in children (Baker-Sennett, Matusov, & Rogoff, 1993; Dreher & Oerter, 1987; Pea & Hawkins, 1987). Both successive refinement and opportunistic planning have been supported empirically, and both appear to explain some central aspect of human planning. Successive refinement models capture the top-down, goal-directed characteristics of human planning (Anderson, 1983) but would lead to the conclusion that young children, who consistently find the use of hierarchies and the process of sequencing difficult, cannot plan (Das et al., 1996). This conclusion has been undermined by results of developmental studies

184


showing that children do plan, and are often quite good at it (see Friedman, Scholnick, & Cocking, 1987 for an excellent review of the development of planning). Opportunistic planning, in turn, has been criticized for making “distractibility” a central aspect of human planning, overlooking a great deal of evidence suggesting that human behaviour is controlled by organized structures (Anderson, 1983). It appears that both views are necessary to accurately describe the process of planning. When factors such as age, cortical damage, knowledge of the planning domain, and constraints placed upon planning efficiency by the human cognitive system are examined, it is clear that both successive refinement and opportunistic planning play a role. Improvements to the analysis and design of total-order planning algorithms (see Figure 9.1) continue to appear in the literature (Blum & Furst, 1997). In the last decade, however, AI research has focused on integrating planning and execution operations with new information becoming known to the agent in the course of action. This work on “reactive” or “dynamic world” planning often uses techniques entirely different from those in the traditional AI planning literature (Chapman, 1991). Reactive planning implies a “least commitment” strategy which allows plans to be more easily modified as planning progresses similar to opportunistic planning. For example, knowledge that there are no newspaper stands beyond the security checkpoint can be integrated into the plan in Figure 9.2 by inserting an arrow from Buy Newspaper to Go Through Security Checkpoint. In contrast, more reasoning and replanning will normally be required to correct a sequential plan relative to new information – the system will have to change ordering decisions to which it has already committed, and will have to reason afresh about the validity of the new plan. The trade-offs for the efficiencies of partial-order representations are that more memory, and in some cases more complex algorithms, are also required.

Figure 9.1 The total planning framework. Unlike opportunistic planning models, the “totalorder” planning framework provides the planning agent with all necessary knowledge from which the agent is expected to produce a complete, fully specified plan before any actions are taken.


185

Figure 9.2 A partially ordered plan. An action at the tail of an arrow must be executed some time before the action of the head of that arrow, but these are the only constraints on execution order. In the shown partial plan, Check Baggage, Go Through Security Checkpoint, and Buy Newspaper may be executed in any order.

COGNITIVE NEUROSCIENCE PERSPECTIVES In general, most cognitive neuroscience (CN) studies have tested subjects in well-structured scenarios where an explicit goal is explained to the subject who then attempts to plan how to achieve the goal before (in some cases) executing the plan (Owen, Downes, Sahakian, Polkey, & Robbins, 1990; Owen, Doyon, Petrides, & Evans, 1996). On occasion, ill-structured problems are presented to the subject who then has to develop the goals of the activity as well as the plan to achieve those goals. In addition, ill-structured problems make it more likely that some form of reactive planning (changing the structure of the plan on-line) will be required during execution of the plan (Spector & Grafman, 1994). A limitation of many CN planning studies is the time domain within which the plan may be developed and executed (see chapter 7 by Owen, chapters 1, 8 by Morris et al., this volume). In most cases, the time is constrained due to the limits of the neuropsychological evaluation procedures and the problem itself is often artificial. In rare instances, the planning and execution of real-life activities are observed where the time scale can be as long as several hours. Although planning problems may also be reflected in the performance of simple everyday tasks like brushing your teeth or making coffee (Schwartz, Reed, Montgomery, Palmer, & Mayer 1991), we focus in this review on higher level cognitive plans (see Box 9.1). The main paradigms used in CN planning studies are those evaluating subject route finding, Tower task performance, and performance in simulated or real-life scenarios. In each of these paradigms, subjects are first usually asked to assemble in their mind the required actions they need to make in order to achieve the instructed goal (planning time) which is followed by the execution of the task (planning execution). Most often, subject response times and accuracy scores are used to derive inferences about planning failures. In general, patients with frontal lobe lesions and those with subcortical disorders (such as Parkinson’s disease, PD) affecting the basal ganglia or cerebellum are the most impaired on a variety of planning paradigms. There are many other CN studies using tasks that require processing similar to what a subject might do in constructing and executing a plan, including script event

186


generation and verification studies whose results we believe are relevant for understanding which brain regions subserve plan processing. Route finding and learning tasks concerned with documenting planning failure require that subjects clearly perceive the geometric properties of the

Box 9.1

A cognitive view of a plan

Based on the discussion of cognitive and computational planning models, it is possible to construct a framework for the cognitive neuroscience investigation of planning. This framework indicates the complexities facing investigators attempting to disambiguate the cause of planning failures in patients or brain activation profiles in normal subjects performing a planning task. Plans have a number of cognitive components that can be seen in the figure in the box. Each plan as well as plan event has a characteristic time duration, event order is generally in a left to right direction although most plans have some level of branching or recursiveness. The number of plan events composing a plan can vary across plans. Plans can be based on well- or ill-structured problems. The key features of plan development are somewhat different from those of plan execution although some overlap is apparent. The general characteristics of plans have to do with conditions for their retrievability and instantiation. Frequency, imagibility, saliency, and motivation are relevant characteristics of any form of knowledge representation including plan-level knowledge. Total-order plans are those that do not allow any deviation from the plan path. Partial-order plans allow for opportunistic deviations and the ability to rejoin the plan path at a juncture close to where you left it. In order to understand the overall meaning of a plan, conceptual and semantic knowledge must be distilled and integrated across individual plan events – each of which has its own semantic and conceptual value. Reactive planning involves the unexpected introduction of another plan event in the main plan path whereas branching is the process whereby the main plan path is appended with a subsidiary path which returns to the main plan path at the point you left it. Well-structured problems have an explicit goal that can induce the plan path selection. Ill-structured problems require the formation of goals and multiple plan paths. Note that the sequential structure of plan events depends to some extent on whether the transition from one plan event to another is characterized by physical (in order to take a shower you must step into the shower stall), social (in the United States, the population generally showers once per day in the morning before eating breakfast and leaving for work), or individual constraints (individuals having their own idiosyncratic sequence in carrying out a plan). This latter constraint or, viewed from a different perspective, ability allows for the most flexible and inventive of human behaviours.


187

route and they have the motoric and attentional skills to perform the task. In order to plan how to navigate through a complex maze, the subject must identify the sequence of proper turns, encode the path as a plan, and commit the plan to memory. The plan execution phase begins as soon as the path is

188


encoded. Patients with prefrontal lesions are able to navigate routes but they are particularly impaired in learning new routes or in utilizing old routes in order to adapt to a new route (Karnath & Wallesch, 1992; Karnath, Wallesch, & Zimmermann, 1991). Flitman and his colleagues have used O15 Positron Emission Tomography (PET) to demonstrate that the right prefrontal cortex is particularly active during the retrieval of an encoded route compared to when a subject is simply traversing a seen maze (Flitman, Cooper, & Grafman, 1997). The anterior cingulate cortex was also active during task performance indicating that subjects actively planned their route and then monitored their chosen path. These findings suggest that subjects tend to navigate mazes by using an opportunistic (rather than a “look-ahead”) strategy which depends more on the immediate maze environment and perceptual/spatial processing whereas the retrieval of a complete cognitive plan for traversing the maze requires prefrontal cortex mediation. Tower-type tasks are composed of three of more pegs with a number of discs sitting on the pegs (see chapter 1 by Ward and Morris, Figures 1.1–1.4). In order to achieve the goal state, subjects have to move the discs and place them on the other pegs according to the rules of the particular tower task adapted. The most commonly used tasks in the CN literature are the Tower of Hanoi (TOH), Tower of London (TOL), and Tower of Toronto (TOT). In general, on Tower tasks, the larger the number of moves required to achieve a goal state, the more difficult the problem appears to the subject. Ward and Allport (1997) using five-disc TOL problems found that planning actions online was limited by subject difficulty in evaluating and selecting one course of action or one subgoal chunk from the set of competing actions at each step in the course of plan execution with increased pre-move preparation time related to the number of competing alternative choices (see also chapter 5 by Ward, this volume). Dehaene and Changeux (1997) used a connectionist model to simulate performance on the TOL task. They postulated that planning requires working memory units, plan units that cause novel activation patterns among lower level operation units generating a plan, and reward units that evaluate the correct or incorrect status of a plan. Each additional indirect move added up to 110 cycles to the stimulation! While humans appear able to chunk an entire action series, Dehaene and Changeux’s model was unable to do so. When they lesioned the plan units in their model, it effectively disconnected the operation network from the reward network, making it difficult for the model to judge the relevance of individual TOL moves to achieving the overall plan. The lesioned model also predicted that a patient might find it difficult to select a move, although they would be able to verify whether a move shown to them would be correct. PET functional neuroimaging studies identify a large set of brain areas which are activated during performance of the TOL by normal subjects. However, when performance on the easy TOL problems was subtracted from


189

the harder TOL problem brain activation profile, only right Brodmann’s area 10 and left Brodmann’s area 9 along with premotor cortex remain significantly activated (Baker et al., 1996). Morris, Ahmed, Syed, and Toone (1993) using SPECT found left prefrontal activation related to TOL performance with the more difficult the problem (as indicated by greater subject planning time and number of moves), the greater the left frontal activation. If normal subjects are given feedback on their TOL performance during PET functional neuroimaging, additional activation can be seen in orbitofrontal and medial caudate brain regions (Elliott, Frith, & Dolan, 1997b). Depressed patients performing the TOL during PET scanning fail to show activation in ventromedial cortex or striatum and show no augmentation in activation from easy to difficult problems (Baker et al., 1997; Elliott et al., 1997a). Goel and Grafman (1995) confirmed that patients with prefrontal cortex lesions are impaired on the TOH task. They noted, however, that the deficit was apparent when patients had to overcome a prepotent strategy and make a counter-intuitive move (regardless of problem difficulty). This finding suggested that frontal lobe lesion patients could initiate a plan similar to controls, but had difficulty branching out from the main plan path (see also Morris, Miotto, Feigenbaum, Bullock, Polkey, 1997). Task-switching capability is important for developing and executing plans when opportunistic shifting between subgoals is necessary. Rogers, Sahakian, Hodges, Polkey, Kennard, and Robbins (1998) found that patients with left frontal lobe lesions showed increased time costs associated with predictable switches between tasks when there was interference between the tasks and when available task cues were relatively weak and arbitrary. Shallice and Burgess (1996, 1991) have studied many patients with frontal lobe lesions as they execute real-life plans. These patients performed normally on many standard cognitive tasks evaluating perception, language, and episodic memory. The real-life tasks involved shopping and similar activities that needed to be performed within a time limit. Their patients were able to remember each task and its different rules of engagement. Impaired plan execution and goal attainment were observed when the patients failed to appropriately divide their time on each of a set of tasks. Some patients appeared unable to reactivate, after a delay, a previously generated intention to perform a task when they were not directly signalled by a stimulus in the environment. Shallice and Burgess speculated that an internal marker may be set for stored plans, whereas for new plans or plan development, the marker(s) may be more fragile and subject to interference when patients have frontal lobe brain damage. They term this deficit a “strategy application disorder”. Goldstein, Bernard, Fenwick, Burgess, & McNeil (1993) evaluated a patient with a unilateral left frontal lesion who performed normally on standard neuropsychological tests and who learned the rules to a Multiple Errands Task (MET) designed by Shallice and Burgess. The patient’s errors

190


were characteristic of a planning failure and included deficits in executing multiple errands, inefficiency (e.g., he could have bought all the goods he needed from one store at one time – instead he returned to the store more than once), rule breaking (he left the neighbourhood to purchase some goods), disinhibition, and post-hoc rationalizations for his inappropriate behaviours. Bechara, Damasio, Damasio, and Anderson (1994) found that patients with ventromedial frontal lobe lesions performing a gambling task tended to choose from the high risk, quick payoff cards rather than the low paying but better long-term risk cards. They hypothesized that these patients had a dissociation between knowledge of the ramifications of their plan and somatic input that would have alerted them to the negative consequences of their chosen plan. Goel, Grafman, Tajik, Gana, and Danto (1997) examined patients with frontal lobe lesions on a realistic financial planning task and found that patients with frontal lobe lesions were impaired at the global level of planning but had normal local level performance. That is, patients with frontal lobe lesions had difficulty in organizing and structuring their plan development space. They were able to begin planning but were unable to adequately divide their cognitive efforts among each planning phase. They spent too much time planning for events that would occur in closer chronological proximity to the planning development time. Patients expressed consternation that there were no right or wrong answers nor obvious termination points in their planning task and tended to terminate the testing session before they specified all their plan details or satisfied the task goals. Interestingly, the patients did not attempt (as did controls) to negotiate some apparent constraints imposed by the investigators. The patients’ planning failures were attributed to difficulty in generalizing from particular events, failure to shift between mental sets, poor judgment regarding the adequacy and completeness of the plan, and inadequate access to structured event complexes (i.e., memory for plan-level attributes such as thematic knowledge, plan grammars, etc.). Scripts are knowledge structures that resemble plans and contain information pertinent to carrying out an action sequence including characterizations of the events, the temporal order of events, and thematic information. Script tasks typically ask participants to sort events, to make decisions about a set of events they are shown (e.g., whether they are in the correct order or belong to the same script), or to carry out a typical script in real time. CN research on script processing (Sirigu, Zalla, Pillon, Grafman, Agid, & Dubois, 1995a, Sirigu, Zalla, Pillon, Grafman, Dubois, & Agid, 1995b, Sirigu, Zalla, Pillon, Grafman, Dubois, & Agid, 1996 in press) indicates that patients with prefrontal cortex lesions have selective difficulty generating (or sorting) an appropriate script event sequence, particularly when the events came from scripts which subjects were less familiar with. Functional


191

neuroimaging findings indicate significant right prefrontal cortex activation when subjects make script event sequence decisions, whereas the left prefrontal cortex is more active when subjects judge whether a single event was a member of a particular script (Partiot, Grafman, Sadato, Flitman, & Wild, 1996). Nichelli, Grafman, Pietrini, Clark, Lee, & Miletich (1995) demonstrated that normal subjects also activated right prefrontal cortex when determining the moral of a story. In addition, when normal subjects mentally generate non-emotional script events, they activate lateral prefrontal and posterior temporal cortices, but when they mentally generate emotional scripts, they activate ventromedial prefrontal and anterior temporal cortices (Partiot et al., 1995). In summary, patients with frontal lobe lesions unambiguously demonstrate difficulty in developing and/or executing a plan. Patients with subcortical lesions to structures that receive frontal lobe projections (e.g., those with Parkinson’s disease or Cerebellar atrophy) may have a similar but more mild planning problem (Berns & Sejnowski, 1998; Grafman, Litvan, Massaquoi, Stewart, Sirigu, & Hallett, 1992; Morris, Downs, Sahakian, Evenden, Heald, & Robbins, 1988; Pascual-Leone et al., 1993; Wallesch, Karnath, Papagno, Zimmermann, Deuschl, & Lucking, 1990). Patients with lesions outside the area of the frontal lobe-basal ganglia-cerebellum axis may fail on planning tasks, not because of an essential deficit to plan-level processes, but because of, for example, cognitive deficits in spatial perception or language comprehension. Category-specific plan impairment may also be observed with ventromedial frontal lobe lesions affecting social cognitive or emotionally arousing plans more than non-social, unemotional cognitive plans (partially due to damage to or dissociation of ventromedial cortex from the autonomic nervous system) (Bechara et al., 1994). Planning processes requiring the structural analysis of plans may be more compromised by left prefrontal lesions whereas the temporal and dynamic aspects of plans may be more compromised by right prefrontal lesions.

CONCLUSIONS To conclude on an optimistic note, sufficient knowledge has been gained from recent CN planning research studies in order to provide some promising leads for future research. The major representation of plan-level knowledge in the human brain appears to be in the prefrontal cortex. The failures in planning associated with prefrontal cortex lesions include problems in both top-down and bottom-up plan development, in the development and execution of novel plans, in the analogical mapping of plans, in parallel processing, in opportunistic/partial-order planning, in time management, in both development and execution of a complete plan event sequence, in discriminating between relevant and irrelevant events, in both well- and

192


ill-structured planning, accessing category-specific plans, and in the adequate successive refinement of plans. The cognitive processing deficits that appear responsible for these planning failures include representational degradation (e.g., making low frequency plans more difficult to retrieve), difficulty in inhibiting pre-potent plans and other action units (disinhibition), deficits in thematic induction (which can hinder plan retrieval), plan grammar deficits (leading to failures in following a sequential path), and modality specific failure in plan development and retrieval (distinguishing between verbal/propositional, visual, and real-time representation of plan behaviour), and impaired opportunistic, partial-order processing. Patients with frontal lobe lesions may exhibit one or more of these deficits. Recent animal and human research suggests that neural networks in the prefrontal cortex are specialized for sustaining information processing over long periods of time, even in the absence of stimulus-specific input (Cohen et al., 1997; Courtney, Ungerleider, Keil, & Haxby, 1997; Fuster, 1995; GoldmanRakic, 1996; Miller, Erickson, & Desimone, 1996). These observations indicate that the neural mechanisms required to support plan-level processing (which by definition would need to occur over long periods of time and needs to be sustained in the absence of stimulus-specific behaviour) are available in the prefrontal cortex. The prefrontal cortex, however, is a large structure and in addition to possessing the general processing capability to handle many aspects of plan-level behaviour including processing multiple plans simultaneously (Lingard & Richards, 1998), the findings we have reviewed also suggest some specificity in the topographical representation of several of the cognitive processes responsible for plans (see Figure 9.3). In this chapter we have argued that the prefrontal cortex is crucial for mediating planning functions that are extended in time and composed of a set of sequential events. It is currently possible to make some claims about the role of several prefrontal cortex regions in planning functions. For this purpose, we can crudely divide the prefrontal cortex into left and right sectors, medial and lateral sectors, dorsal and ventral sectors, and anterior and posterior sectors. Wood and Grafman (2003) have proposed assigning different representational forms of the structured event complex to each of these areas and we can use that same schema for describing planning functions since we believe that plans are just one form of a structured event complex. There is evidence that the left prefrontal cortex focuses on the specific features of individual events (including features and meanings) that make up a plan, whereas the right prefrontal cortex mediates the integration of information across events (including the acquisition of meaning and features at the macro-plan level such as themes and morals). We hypothesize that the medial prefrontal cortex stores key features of predictable overlearned cognitive plans that have a contingent relationship with sensorimotor processes and are


193

Figure 9.3 Mapping planning processes to brain. This cartoon figure of the brain indicates that many plan-level processes are most likely stored and subserved by prefrontal cortex. There is some evidence that indicates which hemisphere (RH = right; LH = left hemisphere) is predominant in mediating a particular planning process (see text). Plan-level knowledge in the prefrontal cortex must be linked or bound to information processing components stored in other brain areas (as shown in the figure) in order for a complete characterization of the plan to be formed. Presumably, the hippocampus and other memory structures contribute to this linkage.

rarely modified. Lateral prefrontal cortex would store key features of plans that are frequently modified to adapt to special circumstances. Ventral prefrontal cortex is concerned with social category specific plans that often have an emotional component. Dorsal prefrontal cortex is concerned more with aspects of plans representing mechanistic activities without a social component (e.g., repairing a food processor). Finally anterior prefrontal cortex tends to represent plans of long duration composed of many events, whereas posterior prefrontal cortex tends to represent plans and actions of short duration and fewer events (e.g., such as a simple association). Since no single prefrontal cortex region would represent all features or components of a plan, specific plans would tend to evoke selected patterns of prefrontal cortex activation. Any region could participate in plan processing depending on the type of plan with the different plan (and cortical) subcomponents being differentially weighted in importance (and activation) depending on the kind of plan, the moment-by-moment demands of the plan, and previous experience with the plan. For example, the left anterior ventromedial prefrontal cortex would be expected to represent a long, multi-event sequence of social interactions (i.e., a social plan) with specialized processing of the meaning and

194


features of single events within the event sequence making up the plan including the computation of their temporal and sequential dependencies and primary meaning. This view differs from that of Newman and colleagues (Newman, Carpenter, & Varma, 2003) who assign different processes to each hemisphere. For example, they hypothesize that the right prefrontal cortex is more involved in the generation of a plan and the left prefrontal cortex is more involved in the execution of a plan based on functional neuroimaging and computational modelling of TOL performance. Our view is that either hemisphere can initiate and guide plan execution, but only specific features of the plans would be encoded and stored in subregions within each hemisphere. The prefrontal cortex is connected to many other brain regions and, in particular, the basal ganglia may play an important role in learning and executing overlearned cognitive plans. We suspect that the execution of a routine plan would rely more and more over time on the simpler sensorimotor components of the activity rather than the associated complex cognitive knowledge contained in the plan. This would result in the recruitment of selected basal ganglia structures to help mediate the sensorimotor activity. In turn this would result in decreased activation in the prefrontal cortex since the cognitive components of the plan would not be explicitly relied upon for its execution. This reduction of prefrontal cortex resources as a plan is routinely repeated allows for the same resources in the prefrontal cortex to be utilized to cognitively focus on another plan leading to the possibility of multitasking behaviour. Many of the subregions within the prefrontal cortex are monosynaptically connected to subcortical structures such as the amygdala and nucleus accumbens that mediate the retrieval of affective input and its binding to cognitive processes, so we anticipate that social plans and other plans of personal relevance to the agent would be infused with emotional cues that bias the development and execution of the plan. Future CN planning research should continue to focus on how event sequences are formed, whether there is unique information that can be abstracted across a sequence of events, whether plans are organized by category and frequency, more explicit depiction of the plan-level cognitive processes referred to in this review (e.g., plan grammars), and more precise mapping of plan components to sectors within the human prefrontal cortex and basal ganglia using patients and functional neuroimaging studies to acquire evidence. It is our view that human CN research is in a position to deliver crucial evidence regarding both the cognitive architecture and the neural topography of plans. By being able to represent and execute plans, we are able to integrate events from the past, present, and future into a single plan-level memory unit (Grafman, 1995; Haith, Benson, Roberts, & Pennington, 1994). Planning enables us to outwit other animals and to cope with a changing environment


195

while achieving myriad goals (Nichelli, Grafman, Pietrini, Alway, Carton, & Miletich, 1994). Planning is the crowning achievement of human cognition and once well understood will enable a complete portrait of the brain’s cognitive functions to be at last attained.

REFERENCES Allen, J. F., Kautz, H. A., Pelavin, R. N., & Tenenberg, J. D. (Eds.). (1991). Reasoning about plans. San Mateo, CA: Morgan Kaufmann. Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press. Baker, S. C., Rogers, R. D., Owen, A. M., Frith, C. D., Dolan, R. J., Frackowiak, R. S. J., et al. (1996). Neural systems engaged by planning: A PET study of the Tower of London task. Neuropsychologia, 34, 515–526. Baker, R., Baker, S. C., Rogers, R. D., O’Leary, D. A., Paykel, E. S., Frith, C. D., et al. (1997). Prefrontal dysfunction in depressed patients performing a complex planning task: A study using positron emission tomography. Psychological Medicine, 27, 931–942. Baker-Sennett, J., Matusov, E., & Rogoff, B. (1993). Planning as a developmental process. In H. W. Reese (Ed.), Advances in child development and behavior (Vol. 24). New York: Academic Press. Bechara, A., Damasio, A. R., Damasio, H., & Anderson, S. W. (1994). Insensitivity to future consequences following damage to human prefrontal cortex. Cognition, 50, 7–15. Berns, G. S., & Sejnowski, T. J. (1998). A computational model of how the basal ganglia produces sequences. Journal of Cognitive Neuroscience, 10, 108–121. Blum, A., & Furst, M. (1997). Fast planning through planning graph analysis. Artificial Intelligence, 90, 281–300. Chapman, D. (1991). Vision, instruction, and action. Cambridge, MA: MIT Press. Cohen, J. D., Perlstein, W. M., Braver, T. S., Nystrom, L. E., Noll, D. C., Jonides, J., et al. (1997). Temporal dynamics of brain activation during a working memory task. Nature, 386, 604–608. Courtney, S. M., Ungerleider, L. G., Keil, K., & Haxby, J. V. (1997). Transient and sustained activity in a distributed neural system for human working memory. Nature, 386, 608–611. Das, J. P., Kar, B. C., Parrila, R. K. (1996). Cognitive planning: The psychological basis of intelligent behavior. New Delhi: Sage. Dehaene, S., & Changeux, J. P. (1997). A hierarchical neuronal network for planning behavior. Proceedings of the National Academy of Sciences, 94, 13293–13298. Dreher, M., & Oerter, R. (1987). Action planning competencies during adolescence and early adulthood. In S. L. Friedman, E. K. Scholnick, & R. R. Cocking (Eds.), Blueprints for thinking: The role of planning in cognitive development (pp. 321–355). Cambridge: Cambridge University Press. Elliott, R., Baker, S. C., Rogers, R. D., O’Leary, D. A., Paykel, E. S., Frith, C. D., et al. (1997a). Prefrontal dysfunction in depressed patients performing a complex planning task: A study using positron emission tomography. Psychological Medicine, 27, 931–942. Elliott, R., Frith, C. D., & Dolan, R. J. (1997b). Differential neural response to positive and negative feedback in planning and guessing tasks. Neuropsychologia, 35, 1395–1404. Fikes, R. E., & Nilsson, N. J. (1971). STRIPS: A new approach to the application of theorem proving to problem solving. Artificial Intelligence, 2, 189–208. Flitman, S., Cooper, V., & Grafman, J. (1997). PET imaging of maze processing. Neuropsychologia, 35, 409–420. Friedman, S. L., Scholnick, E. K., & Cocking, R. R. (Eds.). (1987). Blueprints for thinking: The role of planning in cognitive development. Cambridge: Cambridge University Press.

196


Friedman, S. L., & Scholnick, E. K. (Eds.). (1997). The developmental psychology of planning. Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Fuster, J. M. (1995). Memory and planning: Two temporal perspectives of frontal lobe function. Advances in Neurology, 66, 9–20. Goel, V., & Grafman, J. (1995). Are the frontal lobes implicated in planning functions: Re-interpreting data from the Tower of Hanoi. Neuropsychologia, 33, 623–642. Goel, V., Grafman, J., Tajik, J., Gana, S., & Danto, D. (1997). A study of the performance of frontal patients in a financial planning task. Brain, 120, 1805–1822. Goldman-Rakic, P. S. (1996). The prefrontal landscape: Implications of functional architecture for understanding human mentation and the central executive. Philosophical Transactions of the Royal Society of London, B351, 1445–1453. Goldstein, L. H., Bernard, S., Fenwick, P. B. C., Burgess, P. W., & McNeil, J. (1996). Unilateral frontal lobectomy can produce strategy application disorder. Journal of Neurology Neurosurgery & Psychiatry, 56, 274–276. Grafman, J. (1989). Plans, actions, and mental sets: The role of the frontal lobes. In E. Perecman (Ed.), Integrating theory and practice in clinical neuropsychology. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Grafman, J. (1994). Neuropsychology of higher cognitive processes. In D. Zaidel (Ed.), Handbook of perception and cognition (neuropsychology) (Vol. 15 pp. 159–181). San Diego, CA: Academic Press. Grafman, J. (1995). Similarities and distinctions among models of prefrontal cortical functions. In J. Grafman, K. J. Holyoak, & F. Boller (Eds.), Structure and Function of the Human Prefrontal Cortex (Vol. 769, pp. 337). New York: New York Academy of Sciences. Grafman, J., & Hendler, J. (1991). Planning and the brain. Behavioral & Brain Sciences, 14, 563–564. Grafman, J., Litvan, I., Massaquoi, S., Stewart, M., Sirigu, A., & Hallett, M. (1992). Cognitive planning deficit in patients with cerebellar atrophy. Neurology, 42, 1493–1496. Haith, M. M., Benson, J. B., Roberts, Jr, R. J., & Pennington, B. F. (Eds.), The development of future-oriented processes. Chicago: University of Chicago Press. Hammond, K. (Ed.). (1994). Proceedings of the Second International Conference on Artificial Intelligence Planning Systems. Menlo Park, CA: AAAI Press. Hayes-Roth, B., & Hayes-Roth, F. (1979). A cognitive model of planning. Cognitive Science, 3, 275–310. Hoc, J.-M. (1988). Cognitive psychology of planning. London: Academic Press. Karnath, H. O., & Wallesch, C. W. (1992). Inflexibility of mental planning: A characteristic disorder with prefrontal lobe lesions? Neuropsychologia, 30, 1011–1016. Karnath, H. O., Wallesch, C. W., & Zimmermann, P. (1991). Mental planning and anticipatory processes with acute and chronic frontal lobe lesions: A comparison of maze performance in routine and non-routine situations. Neuropsychologia, 29, 271–290. Kolodner, J. (1993). Case-based reasoning. San Mateo, CA: Morgan Kaufmann. Langley, P., & Drummond, M. (1990). Toward an experimental science of planning. In DARPA (Ed.), Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control (pp. 109–114). Lingard, A. R., & Richards, E. B. (1998). Planning parallel actions. Artificial Intelligence, 99, 261–324. Miller, E. K., Erickson, C. A., & Desimone, R. (1996). Neural mechanisms of visual working memory in prefrontal cortex of the macaque. Journal of Neuroscience, 16, 5154–5167. Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. New York: Holt, Rinehart & Winston. Morris, R. G., Downes, J. J., Sahakian, B. J., Evenden, J. L., Heald, A., & Robbins, T. W. (1998).


197

Planning and spatial working memory in Parkinson’s disease. Journal of Neurology, Neurosurgery, & Psychiatry, 51, 757–766. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, R. K. (1993). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367–1378. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., Polkey, C. E. (1997). The effect of goal–subgoal conflict on planning ability after frontal- and temporal-lobe lesions in humans. Neuropsychologia, 35, 1147–1157. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Newman, S. D., Carpenter, P. A., Varma, S., & Just, M. A. Frontal and parietal participation in problem solving in the Tower of London: fMRI and computational modeling of planning and high-level perception. Neuropsychologia, 41, 1668–1682. Nichelli, P., Grafman, J., Pietrini, P., Alway, D., Carton, J. C., & Miletich, R. (1994). Brain activation during chess deliberation. Nature, 369, 191. Nichelli, P., Grafman, J., Pietrini, P., Clark, K., Lee, K. Y., & Miletich, R. (1995). Where the brain appreciates the moral of a story. Neuroreport, 6, 2309–2313. Owen, A. M. (1997). Cognitive planning in humans: Neuropsychological, neuroanatomical and neuropharmacological perspectives. Progress in Neurobiology, 53, 431–450. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. Planning and spatial working memory following frontal lobe lesions in man. Neuropsychologia, 28, 1021–1034. Owen, A. M., Doyon, J., Petrides, M., & Evans, A. C. (1996). Planning and spatial working memory: A positron emission tomography study in humans. European Journal of Neuroscience, 8, 353–364. Partiot, A., Grafman, J., Sadato, N., Wachs, J., & Hallett, M. (1995). Brain activation during the generation of non-emotional and emotional plans. Neuroreport, 6, 1269–1272. Partiot, A., Grafman, J., Sadato, N., Flitman, S., & Wild, K. (1996). Brain activation during script event processing. Neuroreport, 7, 761–766. Pascual-Leone, A., Grafman, J., Clark, K., Stewart, M., Massaquoi, S., Lou, J.-L., et al. (1993). Procedural learning in Parkinson’s disease and cerebellar degeneration. Annals of Neurology, 34, 594–602. Pea, R. D., & Hawkins, J. (1987). Planning in a chore-scheduling task. In S. L. Friedman, E. K. Scholnick, & R. R. Cocking (Eds.). Blueprints for thinking: The role of planning in cognitive development (pp. 273–302). Cambridge: Cambridge University Press. Rogers, R. D., Sahakian, B. J., Hodges, J. R., Polkey, C. E., Kennard, C., & Robbins, T. W. (1998). Dissociating executive mechanisms of task control following frontal lobe damage and Parkinson’s disease. Brain, 121, 815–842. Rosenbloom, P. S., Laird, J. E., & Newell, A. (Eds.). (1993). The SOAR papers: Research on integrated intelligence. Cambridge, MA: MIT Press. Sacerdoti, E. D. (1975) The nonlinear nature of plans. Advance Papers of the Fourth International Joint Conference on Artificial Intelligence, 206–214. Schank, R. C., Abelson, R. P. (1997). Scripts, plans, goals, and understanding. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Scholnick, E. K., Friedman, S. L. (1987). The planning construct in the psychological literature. In S. L. Friedman, E. K. Scholnick, & R. R. Cocking (Eds.), Blueprints for thinking: The role of planning in cognitive development (pp. 3–38). Cambridge: Cambridge University Press. Schwartz, M. F., Reed, E. S., Montgomery, M., Palmer, C., & Mayer, N. H. (1991). The quantitative description of action disorganization after brain damage: A case study. Cognitive Neuropsychology, 8, 381–414. Shallice, T. & Burgess, P. (1996). The domain of supervisory processes and temporal organization of behaviour. Philosophical Transactions of the Royal Society of London, B351, 1405–1412.

198


Shallice, T., & Burgess, P. W. Deficits in strategy application following frontal lobe damage in man. Brain, 114, 727–741. Sirigu, A., Zalla, T., Pillon, B., Grafman, J., Agid, Y., & Dubois, B. (1995a). Selective impairments in managerial knowledge following prefrontal cortex damage. Cortex, 31, 301–316. Sirigu, A., Zalla, T., Pillon, B., Grafman, J., Dubois, B., & Agid, Y. (1995b). Planning and script analysis following pre-frontal lobe lesions. In J. Grafman, K. J. Holyoak & F. Boller (Eds.). Structure and function of the human prefrontal cortex (Vol. 769, pp. 277–288). New York: New York Academy of Sciences. Sirigu, A., Zalla, T., Pillon, B., Grafman, J., Dubois, B., & Agid, Y. (1996). Encoding of sequence and boundaries of scripts following prefrontal lesions. Cortex, 32, 297–310. Sirigu, A., Cohen, L., Zalla, T., Pradat-Diehl, P., Van Eeckhout, P., Grafman, J., et al. (in press). Distinct prefrontal regions for processing sentence syntax and story grammar. Cortex. Spector, L., & Grafman, J. (1994). Planning, neuropsychology, and artificial intelligence: Cross fertilization. In F. Boller & J. Grafman (Eds.), Handbook of neuropsychology (Vol. 9, pp. 377–392). Amsterdam: Elsevier Science Publishers. Suchman, L. A. (1987). Plans and situated actions. Cambridge: Cambridge University Press. Wallesch, C.-W., Karnath, H. O., Papagno, C., Zimmermann, P., Deuschl, G., & Lucking, C. H. (1990). Parkinson’s disease patient’s behavior in a covered maze learning task. Neuropsychologia, 28, 839–849. Ward, G., & Allport, D. A. (1997). Planning and problem-solving using the five-disk Tower of London task. Quarterly Journal of Experimental Psychology, 50A, 49–78. Wilensky, R. (1983). Planning and understanding: A computational approach to human reasoning. Reading, MA: Addison-Wesley. Wood, J. N., & Grafman, J. (2003). Human prefrontal cortex: Processing and representational perspectives. Nature Reviews Neuroscience, 4, 139–147.

CHAPTER TEN

The search for specific planning processes Paul Burgess Institute of Cognitive Neuroscience and Department of Psychology, University College London, London

Jon S. Simons Institute of Cognitive Neuroscience and Department of Psychology, University College London, London

Laure M.-A. Coates Department of Clinical Psychology, Institute of Psychiatry, London

Shelley Channon Department of Psychology, University College London, London

INTRODUCTION This chapter is based on an argument first advanced at a symposium on the psychology of planning, held at the University of Kent, UK in September 2002. As part of this symposium, one of us (PB) presented evidence which questioned the assumption that there may be specific processes in the brain devoted to an activity termed “planning”. After the presentation, one of the other delegates (also an author of a chapter in the present volume) jokingly suggested that the talk should have been called “There Are No Planning Processes”. This was a deliberately ludicrous extension of the argument: there is little doubt that people do make plans for the future (e.g., deciding what they will do at the weekend; planning a holiday, etc.), and that this mental activity must have correspondence in the brain. So in that sense there must be “planning processes”. However, there was also little doubt that the jibe nicely summed up an important point theme of the argument. This was: (a) the question of whether there are processes that are invariably used in planning and not in other activities in the way that, say, visual object 199

200

BURGESS, SIMONS, COATES, CHANNON

recognition systems are currently thought of as involved only in the visual recognition of objects (b) if such processes do exist, is the present course of enquiry that we are following likely to discover them? The preponderance of the term “planning processes” in the literature, and the dominance of the use of one particular procedure for study (the Tower of London (TOL), Shallice, 1982, and its variants), suggests that (a) is a widely held assumption. However, this assumption has only rarely been challenged. This chapter aims to provide such a challenge, and to outline ways in which the focus of planning research in the field of cognitive neuroscience needs to be broadened.

CHALLENGE 1: IS “PLANNING” JUST A LABEL FOR A RANGE OF DISPARATE HUMAN ACTIVITIES? In Professor Grafman’s excellent chapter (9) in this volume, he uses the term “ubiquitous” to describe the activity of “planning”. This suggests two possibilities. The first is that planning is a good candidate for study by cognitive neuroscientific methods since it predicts that: • there will be many situations which could be said to be cardinal of planning, making experimentation and generalization of results easier • the relevance of findings to an understanding of how the brain supports planning activity would be considerable. On the other hand, however, it could suggest a second view. This is that the term is just a label for a whole set of activities which actually share very few cognitive processing resources; in other words that the term has poor “construct1 validity”. This is not so unlikely: our language is replete with such labels (e.g., “travelling”, which could cover movement by bus, train, walking,

1

The term “construct” here refers to a hypothetical set of processes or abilities which support a given activity. Examples abound in cognitive neuroscience (“episodic memory” and “working memory” are two of the most frequently encountered). Formally, a construct has “indicators”, which are sets of data which are considered evidence for the existence of it. In the current context these might be scores from neuropsychological tests. For instance, one might suggest that performance on a word list free recall paradigm would be an indicator for an “episodic memory” construct. Another might be, say, recollection of what one had for breakfast this morning. The two tasks (recall of a word list and remembering what one had for breakfast) are quite different, but one assumes that they share some “episodic memory” processing resources. In this way, two strong indicators for a construct should generally show a greater performance correlation with each other than occurs between one of them and another task which supposedly has no “episodic memory” component.

10. THE SEARCH FOR SPECIFIC PLANNING PROCESSES

201

sailing, etc.), and the brain is unlikely to be conveniently organized according to our linguistic principles. The first possibility is largely predicated on the view that the essential feature of “planning” is a function which we might call “look-ahead” (e.g., McCarthy & Warrington, 1990). “Look-ahead” in this context refers to the function of working out in one’s mind every stage between the present state and some desired outcome. It assumes a clear articulation of the goal state, and a logical progression (by means–end analysis or similar) of determining substeps towards that goal. In this view of planning, every substep should be a progression of the last, and deviations from this path would be seen as failures. The aim is to produce the most pragmatic, logical and economical path to the intended goal. However, how often do people actually do this in everyday life? One can of course construct experimental tasks which are so constrained that virtually the only way to progress is in this manner. However, unless this is how people normally choose to plan, the results are likely to be of little practical relevance to understanding how people behave, and how the brain supports that behaviour. In fact we could go further, and entertain the thought that planning in the sense that it is conceived of by psychologists (in other words as an activity with a core defining feature, namely “look-ahead”) is only a minor aspect of cognition, in the sense that it is only used in very specific circumstances and is not central to the activity that most people would identify as “planning”. We will investigate this possibility later in the present chapter through a review of the relevant experimental literature, and examples of people actually planning real-world type activities. However, there is first an important point to be made about the experimental status of “look-ahead” processes, even if they do exist. This is that it is difficult to distinguish, behaviourally, between the product of the operation of the putative ability (“look-ahead”) and the application of the product of that operation some time distant. In other words, we do not make our planning decisions in a vacuum, devoid of all knowledge of what we previously decided to do in similar situations, or where similar goals were trying to be achieved. In a new situation, or where a new goal has to be reached, perhaps “look-ahead” may be used to come to determine a course of action (i.e., a plan). However, when one subsequently encounters a similar situation/goal again, one now has the option of referring to what one did before. We will refer to these products of previous planning episodes as “stored preferences” (“stored” because one can express a novel preference in new situation, and this refers instead to remembering past decisions; “preferences” because they are “guides” to behaviour rather than absolute rules). We will describe in more detail what we mean by “stored preferences” shortly. This crucial distinction between novel planning and the expression of stored preferences is

202


an important part of our argument, and will be returned to throughout this chapter. However, first we will give an everyday example as an illustration.

CHALLENGE 2: PLANNING AS THE EXPRESSION OF STORED PREFERENCES In the summer of 2001, one of us (PB) was having a discussion with Professors Tim Shallice and Donald T. Stuss. The topic of discussion was the processing demands of everyday life situations, and the example we were discussing was making grocery purchases in a supermarket. Both professors pointed out that they habitually checked and compared the prices of different brands before they bought them. In fact they would occasionally go as far as checking the weight or quantity of the packet, and making a brief assessment of its relative value compared with some other size or volume. Often they would plan where to go on the basis of their knowledge of relative prices. Moreover, when in a particular place for some other reason, they might consider making local purchases in advance of when the food would be consumed, given that it made financial and practical sense. Clearly the formation and execution of strategies to minimize expense and effort that they were describing constitute “planning” at some level: They had a goal state which they wished to achieve (full stomachs) and many constraints they had to negotiate (spending as little money as possible; buying food that they could cook which would meet the approval of their own and others’ tastes; the availability both present and in the future of various ingredients; storage of the food and its perishability in relation to their weekly schedule and so forth). However, PB was unsettled by this discussion since he never indulged in this behaviour: Perhaps this meant that he had no planning ability. There is an alternative explanation. Some time ago PB had cause, when particularly hard up, to assess the minutiae of his weekly outgoing expenses. This revealed that only a small fraction of his cost of living was spent on food. Accordingly, the amount of money that he could have saved by buying the cheapest products in the largest economy-size amounts was negligible compared with the savings to be made elsewhere. So he decided that there was little point in looking to this area of his life to make savings, and since that day has very rarely checked the prices of foodstuffs. Now consider the situation where as a conclusion of this discussion between these colleagues it was decided to use as a test of planning ability a measure derived from observations of people’s strategies and behaviour relating to food purchases. Had PB been a subject in such a study, he would have scored zero on this measure since he would have shown little or no evidence of planning, and the experimenter might have concluded that PB had limited planning ability. Of course that might be the case. However, it is doubtful that this measure would have demonstrated it. The decision not to indulge in the


203

planning behaviour was the result of much careful planning and consideration many years previously. The accurate measure of PB’s planning ability would have to have been made at the point of that original decision. Let us now imagine that the study had included a large number of people, half of whom behaved as Professors Shallice and Stuss, and half behaving as PB does (i.e., never checking prices), and the results of the measure were correlated with the groups’ performances on a range of psychometric tests of memory, executive function and so forth. What might the results be? Very low correlations with the planning tasks (and between the planning tasks), since the behavioural measure for the “never check prices” group would not be an accurate measure of potential planning ability. In other words the behavioural indicator has poor “construct validity”. We will return to this point later. It is not fanciful to suggest that many situations one might conceive of as requiring planning may be no different from this example. Perhaps what we call planning is for a large part the expression of preferences based on prior experience and previous consideration, and the overlap in processes that are recruited will differ radically from one individual to the next. Let us give a practical example of what these “preferences” might be, and how they might affect the planning process. “Preferences” here can refer to a range of beliefs or experience-based “rules” that impinge upon choice of behaviour. The beliefs might be abstract such as “men don’t like shopping”, or person specific as in “I wouldn’t like shopping if I tried it”. Or they might be experienced-based as in: “It’s not worth travelling miles to my out-of-town supermarket when I can get the same products for just a few pence more locally,” “Preferences” can of course also refer to quite simple determinants of choices such as “I prefer oranges to apples”, but it is the larger determinants that we are principally concerned with here. One can see the relevance to the activity called “planning” if one considers how one actually solves the problem of needing to buy some food. How much “look-ahead” is required here, and how much is one’s behaviour determined by one’s stored “preferences”? Of course a retort to this argument might be that the real “executive” component of planning concerns solving a problem where one does not have a ready solution (cf. novelty). For instance, consider the present example if one moves to a different town, where it is necessary to discover new sources of food. However, in that circumstance one’s set of stored preferences is still likely to be a determinant of the course of action selected (e.g., finding a local shop vs. a supermarket). If “planning” is the activity by which we determine a course of action that takes us towards a desired state, then not to study the influence of stored preferences (and the way they are formed) will surely leave us understanding only a small part of the activity. Our view is that whilst “look-ahead” might be one processes involved in determining a future course of action, it is likely to be a relatively small part that is used in only specific circumstances. The eventual “plan” will

204


in fact be the product of other processing: how someone conceptualizes their current situation, and what they notice about it; their needs and priorities; the success of past actions and so forth. According to this conceptualization of planning, the most important aspect to study would be how the brain supports these other activities, rather than concentrating on the relatively small component of “look-ahead” as is currently the vogue.

Experimental consequences This argument introduces the possibility that “look-ahead” might only be a small determinant of “planning” behaviour. It is likely that the translation of stored preferences into an intended course of action would be a complex matter that would use many different processing resources. For instance, obvious candidates would be the assigning of relative importance to potential outcomes, and determining the personal cost (in time, effort, pleasure, etc.) of achieving them. The results of this kind of analysis are likely to be quite idiosyncratic. If this analysis is correct, then planning may involve many more executive control processes than merely “look-ahead” (e.g., those involved in inhibition, autobiographical recollection, assessment of reward, etc.). Let us now consider whether the experimental evidence is consistent with these views. We will consider first evidence from other laboratories, and then describe some recent work of our own.

CHALLENGE 3: IS THE EXISTING EXPERIMENTAL EVIDENCE CONSISTENT WITH THE ASSUMPTION THAT “LOOK-AHEAD” IS THE PRINCIPAL CONSTRUCT UNDERPINNING PLANNING PERFORMANCE? The emerging field of artificial intelligence (AI) in the 1970s provided the first impetus for experimental studies of planning, with the introduction of a particular task, the Tower of Hanoi (TOH), to study problem solving and planning (Simon, 1975). This task was useful to study because it appeared amenable to means–end type analysis by presenting the subject with a very highly constrained situation with limited response options and little opportunity for the application of the kinds of idiosyncratic preferences that actually govern people’s choices in the real world. Subsequently, a variant called the TOL test (invented by Shallice & McCarthy; see Shallice, 1982), was adopted by cognitive psychologists and neuropsychologists in the 1980s to study the cognitive processes necessary for successful planning. The TOH and TOL are similar in nature, although not identical. Common properties include the requirement to rearrange a number of discs or beads on three pegs until a specified goal arrangement is reached, with the constraint that only one disc – the topmost one on any peg – may be moved at a time. Where


205

the tasks differ is that in the TOH the discs are differently sized, with the additional constraint that a smaller disc cannot be moved on top of a larger one.

Studies involving the Tower tasks The Tower paradigms have been used to assess planning abilities in numerous neuropsychological patient groups, such as individuals with focal brain lesions (Carlin, Bonerba, Phipps, Alexander, Shapiro, & Grafman, 2000; Goel & Grafman, 1995; Morris, Miotto, Feigenbaum, Bullock, & Polkey, 1997; Owen, Downes, Sahakian, Polkey, & Robbins, 1990; Shallice, 1982), closed head injury (Levin, Mendelsohn, Lilly, & Fletcher, 1994), medial temporal lobe amnesia (Butters, Wolfe, Martone, Granholm, & Cermak, 1985; Xu & Corkin, 2001), temporal lobectomy (Morris et al., 1997), frontotemporal dementia (Carlin et al., 2000; Simons, Verfaellie, Galton, Miller, Hodges, & Graham, 2002), cerebellar atrophy (Grafman, Litvan, Massaquoi, & Stewart, 1992), Huntington’s disease (Butters et al., 1985), Parkinson’s disease (Morris, Downes, Sahakian, Evenden, Heald, & Robbins, 1988; Owen et al., 1992; Owen, Doyon, Dagher, Sadikot, & Evans, 1998), schizophrenia (Goldberg, Saint-Cyr, & Weinberger, 1990; Morris, Rushe, Woodruff, & Murray, 1995), obsessive compulsive disorder (Veale, Sahakian, Owen, & Marks, 1996), depression (Elliott, Sahakian, Michael, Paykel, & Dolan, 1998), and autism (Hughes, Robbins, & Russell, 1994). Probably the most instructive sets of studies for the present argument, however, concern those that involve patients with lesions restricted to the prefrontal cortex since the possible reasons for their failure on the task are likely to be fewer than for pathologies which may affect many brain regions. These studies have typically documented impairment compared with controls in various aspects of performance. The commonest findings show that patients require more moves and more time to achieve solutions (Carlin et al., 2000; Goel & Grafman, 1995; Morris et al., 1997; Owen et al., 1990; Shallice, 1982; although see Shallice, 1988). Interestingly, however, these impairments have often been observed in the context of normal initial thinking (planning) time, which has led some researchers to suggest that the difficulties experienced by prefrontal lesion patients on these tasks may reside mainly in processes related to the execution of plans (e.g., Carlin et al., 2000) rather than the formation of the plans themselves. We will call the conclusion from this finding, Conclusion 1. Conclusion 1. People with prefrontal damage may fail “planning tasks” because of their inability to follow their plans rather than formulate them.

206


Evidence from functional neuroimaging studies involving healthy volunteers has also highlighted the important role played by regions of prefrontal cortex during performance of Tower tasks (Baker et al., 1996; Dagher, Owen, Boecker, & Brooks, 1999; Dagher, Owen, Boecker, & Brooks, 2001; Morris, Ahmed, Syed, & Toone, 1993; Owen, Doyon, Petrides, & Evans, 1996; Rowe, Owen, Johnsrude, & Passingham, 2001). More particularly, several neuroimaging studies have taken the view that the role of frontal lobe areas such as the dorsolateral prefrontal cortex (DLPFC) is related to the complexity of the problem and therefore the amount of initial thinking time involved (Dagher et al., 1999; Dagher et al., 2001; Morris et al., 1993; Rowe et al., 2001). We will call the conclusion from this finding Conclusion 2. Conclusion 2. Functional imaging studies of Tower tasks suggest complexity of the problem to be an important factor in determining DLPFC results.

Planning or goal–subgoal conflict resolution? How might this apparent disagreement be resolved? Some researchers have noted that the difficulties experienced by some patients on the Tower tasks and related paradigms are especially apparent on those problems that require for their solution a counterintuitive backward move which, at first glance, appears to be leading away from the goal (Colvin, Dunbar, & Grafman, 2001; Goel & Grafman, 1995; Morris et al., 1997). Goel and Grafman (1995) suggested that frontal lobe patients have particular difficulty in identifying and resolving these goal–subgoal conflicts. Support for this view came from a study by Morris et al. (1997), who found that patients with left frontal lobe lesions were impaired selectively on those TOH problems that contained goal–subgoal conflicts. They argued that in the TOH, a goal–subgoal conflict might represent a situation where frontal lobe patients may be unable to inhibit the prepotent response to select a move that appears ostensibly to be leading towards the goal (Morris et al., 1997). We will call the conclusion from this finding Conclusion 3. Conclusion 3. People with frontal lobe damage may fail only certain aspects of Tower tasks: those involving goal–subgoal conflicts.

On the basis of such observations, Goel and Grafman (1995) persuasively argued that impaired performance on tasks such as the TOH and TOL could be explained as one symptom of a generalized response inhibition deficit, without requiring recourse to an interpretation in terms of impaired “planning” functions (i.e., because the inhibition problem would mean that they did


207

not stop and consider the situation appropriately). Furthermore, according to Goel and Grafman, “look-ahead” abilities are neither necessary to solve Tower problems, because the strategies used by subjects can be implemented in computer simulations which do not look-ahead several moves, nor are they sufficient for successful performance, because looking ahead will not on its own resolve goal–subgoal conflicts. We will call the conclusions from these findings Conclusions 4 and 5. Conclusion 4.

Tower-type tasks may not require looking ahead.

Conclusion 5. The aspects of Tower tasks with which frontal patients have difficulty require functions other than just looking ahead.

Goel and his colleagues have also pointed out that Tower tasks are not very realistic examples of planning problems since they are “well-structured” problems, in which the start state and end goal state are both clearly defined, and the rules and constraints of the task are specified in advance (Goel, Grafman, Tajik, Gana, & Danto, 1997; Goel, Pullara, & Grafman, 2001). However, the kinds of real-world situations that patients with prefrontal cortex damage often have difficulties with are typically “ill-structured” in nature, such that start and goal states may be poorly defined, and the rules for undertaking the task, as well as the markers of success, may need to be inferred by the participant. We will call the conclusion from this finding Conclusion 6. Conclusion 6. Tower tasks do not present planning situations that are much like real-world planning situations, and therefore probably make different cognitive demands.

Moving beyond the Tower tasks Several recent studies have attempted to go beyond traditional formulations and to provide a more detailed analysis of the cognitive processes that may be involved in planning abilities. Dehaene and Changeux (1997) utilized a neural network implementation of the TOL which contained three hierarchical levels of processing and could solve TOL problems of varying complexity. When the model was lesioned, by removing the highest level of processing units, the model failed in similar ways to patients with prefrontal cortex damage, with inability to guide the selection of motor operations by evaluation of their relevance to reaching the goal. A recent functional neuroimaging study used a similar cognitive modelling framework to generate expected haemodynamic response functions for different aspects of task performance (Fincham, Carter, van Veen, Stenger, & Anderson, 2002). The authors

208


reported that parametric variations in planning load were associated with differential engagement of dorsolateral prefrontal regions, as well as premotor and parietal cortices. However, more direct experimental evidence was presented by Rowe et al. (2001) who observed activation in dorsolateral prefrontal cortex when participants were both planning moves towards a goal and planning moves unconstrained by a goal. This result suggests that the involvement of dorsolateral prefrontal cortex in planning may be related to generating, selecting, and/or remembering mental moves, but not necessarily to the evaluation of these moves in relation to a specified goal. We will call the conclusion from this finding Conclusion 7. Conclusion 7. DLPFC involvement in “planning tasks” may not reflect the operation of processes concerned with evaluation of steps towards the goal. Using a different paradigm, Koechlin, Corrado, Pietrini, & Grafman (1999) used functional neuroimaging techniques to identify the regions of prefrontal cortex involved in various aspects of planning behaviour. When undertaken separately, successively alternating between two tasks and holding a goal representation in mind over time were both associated with activation in bilateral dorsolateral prefrontal cortex, middle frontal gyrus, and lateral parietal cortex. The interaction between these two processes, characterized as holding a goal in mind while processing a secondary goal, which Koechlin et al. termed “cognitive branching” and suggested to be required during planning, was associated with activation only in frontopolar cortex regions previously implicated in multitasking disorders involving poor plan following (Burgess, Veitch, de Lacy Costello, & Shallice, 2000; Shallice & Burgess, 1991). In more recent work, Koechlin and colleagues provided evidence that regions of frontopolar cortex could be functionally dissociated, with the medial aspect involved in processing when intermediate tasks occur in expected sequences, and the lateral aspect implicated when intermediate tasks occur in unpredictable sequences (Koechlin et al., 2000). This is consistent with the recently proposed role for lateral and medial rostral prefrontal cortex in prospective memory paradigms, in which participants perform one task while keeping in mind a second task that is to be performed on the appearance of a trigger, which typically occurs unpredictably (Burgess, Quayle, & Frith, 2001; Burgess, Scott & Frith, 2003). In this way, for Koechlin, “planning” now encompasses simple expectations of what might happen rather than deliberate plans of intended actions. We will call the conclusion from this finding Conclusion 8. Conclusion 8. For some theorists “planning” involves simple expectations of what might happen rather than complex plans of intended actions.


209

Real-world planning situations The recognition that most laboratory tests of “planning” ability, such as the TOH and TOL, may actually assess different cognitive processes from those necessary for performance in real-world planning situations was reflected in the call from Shallice and Burgess (1991) for the development of “quantifiable analogues of . . . open-ended multiple subgoal situations” similar to those that are encountered in everyday life (p. 728). A number of recent studies have attempted to utilize such analogues of real-world planning situations in studies of patients with prefrontal damage, and have provided further insights into the brain regions necessary for successful planning abilities. For example, Miotto and Morris (1998), using a Virtual Planning Test (VPT) which required the planning and execution of a set of target activities that were to take place over a four-day period, found that patients with frontal lobe lesions tended to select inappropriate activities in relation to the current context. We will call the conclusion from this finding Conclusion 9. Conclusion 9. The root of some frontal lobe patients’ problems in planning situations is in the selection of inappropriate activities. In a similar virtual environment, Zalla, Plassiart, Pillon, Grafman, and Sirigu (2001) examined the ability of frontal lobe patients to formulate and then execute plans of action. The patients took longer than controls to formulate plans, but there were few differences between groups in terms of the nature of the plans generated. Deficits were observed in executing plans, however, which Zalla et al. characterized as action slips, omissions of steps, failure in initiating actions, and purposeless movements between locations. This result echoes Conclusion 1 above, that individuals with prefrontal damage may have greater difficulty following plans than formulating them. Some of the most interesting studies of ill-structured planning situations were conducted by Goel and Grafman (2000; Goel et al., 1997), who used verbal protocol analysis to quantify and analyse planning behaviours of patients with frontal lobe lesions on real-world planning tasks. Goel et al. (1997) examined performance on a financial planning task, in which participants had to achieve four financial goals by manipulating various budgetary variables such as income, expenditure, and the allocation of assets. The patients exhibited difficulty in formulating and structuring their plans of action, which Goel et al. attributed primarily to inadequate access to socalled “structured event complexes” which guide routine behaviour, due to damage either to control mechanisms or to the knowledge representations themselves. The patients also had difficulty with executing their plans, exhibiting poor judgement about how well they were proceeding and when they should stop.

210


This finding was supported by Channon and Crawford (1999), who examined real-life-type problem solving and found that, relative to the control group, those with frontal lobe lesions showed difficulties in targetting the most relevant aspects of the problem situation when generating solutions. Another study (Channon & Crawford, unpublished data) examined planning using real-life-type materials, and found that those with frontal lesions were poorer than controls in prioritizing tasks that were more urgent or had greater consequences if left undone. Similar results were found in a single case study of a professional architect with a right prefrontal lesion who was tested on a real-world architectural design/planning task (Goel & Grafman, 2000). Compared with a matched control, the patient exhibited difficulties in the preliminary design phase, in moving from structuring to solving the problem, and in developing, refining, and detailing design elements. Goel and Grafman characterized his deficit in terms of dysfunction in the lateral transformation of one idea into another. They suggested that such lateral transformation of concepts represents one of the key elements of ill-structured problem solving, and postulated that the right dorsolateral prefrontal cortex may subserve the cognitive operations involved in such processing (we return to this point later). We will call the conclusion from this finding Conclusion 10. Conclusion 10. The reason for frontal lobe patients’ problems on real-world type planning tasks appears to be related more to the creation and selection of appropriate solutions, and the construction of them into a coherent whole, than to a deficit in “look-ahead” ability per se.

SUMMARY SO FAR Most of the planning studies in cognitive neuroscience have used Tower tasks. This is because since Shallice’s (1982) influential paper, it has been assumed that (a) Tower tasks are good models of the planning component involved in more real-world tasks; (b) they are relatively pure measures of a construct (i.e., “look-ahead”). However, 20 years on both these assumptions are challenged by the experimental evidence that has since accumulated. It is now doubtful that Tower tasks are good models for planning in more realistic situations because they are too “structured” (i.e., there are limited choices of possible action, with clearly defined rules and constraints). In contrast, it seems likely that the greatest determinant of performance in everyday life situations that require planning is in fact how well we can deal with the “ill-structuredness” of the situation. Moreover, it seems unlikely that Tower tasks are in any case good measures of “look-ahead” even if this were a large part of the activity we call “planning”. Accordingly, it is doubtful that people with frontal lobe damage principally fail such tasks because of an inability to


211

“look-ahead”. Instead they may fail for many other reasons, including specific problems with resolution of goal–subgoal conflicts, the selection of inappropriate (e.g., rule-breaking) or irrelevant (e.g., discursive) actions, and problems in effecting their intended plan (which may have many subcategories (e.g., action lapses, prospective memory failures, etc.). If this were not enough, there is even debate concerning the most appropriate interpretation of the consistent relation that has been observed, using various methods, between DLPFC function and planning. It is possible that these findings reflect a function which, whilst central to Tower-type tasks, is certainly not exclusive to it, and perhaps plays only a minimal part in other forms of planning task (an obvious candidate might be short-term rehearsal or “working memory”).

EXPERIMENTAL EVIDENCE FROM OUR LABORATORY We will now describe some experimental evidence from our laboratory, where we have been studying (like the others mentioned above) planning in more realistic situations. Our findings not only support the conclusion from the experimental evidence just outlined, but extend them. In particular they emphasize the importance of “stored preferences” in real-world situations. The first thing to note about our results is that Professor Grafman’s comment about planning being a ubiquitous activity is borne out by the frequency with which planning deficits are noticed in neurological patients. Consider Table 10.1. This shows the levels of reporting of the 20 commonest dysexecutive symptoms in a mixed aetiology neurological group of 92 individuals. Carers and relatives of the patients were asked to rate the severity of each of the problems in the patients they knew well, and a comparable questionnaire (the DEX Questionnaire) was given to the patients for them to rate their own problems. In this analysis only ratings of 3 or 4 (out of a maximum of 4) for each item on the DEX questionnaire (Burgess, Alderman, Emslie, Evans, & Wilson, 1996) were considered as indicating a problem. These correspond to classification of the symptom as “often” or “very often” observed. Table 10.1 clearly shows that of all dysexecutive symptoms, planning problems are the ones most often noted by carers and relatives as a problem with, quite astonishingly, almost half of them noticing the symptom “often” or “very often”. Moreover, one can see that planning problems show the greatest disagreement in ratings between the relatives/carers and the patients. Planning problems occurred more frequently than other dysexecutive symptoms according to carers (48% against a background of mean symptom reporting of 27%), but slightly less frequently than others according to the patients (16% against a mean of 17.4%) (for further discussion on this point see Burgess & Robertson, 2002). Clearly planning problems are indeed observed remarkably frequently in neurological patients, as is underreporting of them by the patients.

212


TABLE 10.1 Frequency of reporting of 20 of the commonest dysexecutive problems in a mixed aetiology neurological group (adapted from Burgess & Robertson, 2002; data originally from the study by Wilson et al., 1996) Symptom

Carers %

Patients %

Planning problems

48

16

Distractibility

42

32

Lack of insight

39

17

Poor decision making

38

26

Social unconcern

38

13

Euphoria

28

14

Restlessness

28

25

Apathy

27

20

Lack of concern for others’ feelings

26

26

Perseveration

26

17

Aggression

25

12

Temporal sequencing problems

25

18

Social disinhibition

23

15

Shallow affect

23

14

Impulsivity

22

22

Response inhibition problems

21

11

Poor abstract thought

21

17

Knowing–doing dissociation

21

13

Variable motivation

15

13

5

5

Confabulation

These results underline, at a practical level, the importance of understanding how planning is performed, and the form that planning deficits can take. However, at a theoretical level they may also support the view that many different cognitive processes are involved in planning, given that planning impairments appear to be prevalent in such a wide range of pathologies. Contrast the frequency of planning problems with, say, confabulation, in Table 10.1. Confabulation has been argued to have a quite specific information-processing basis in those who do not show it in the context of general confusion (Burgess & Shallice, 1996) and a circumscribed (though not necessarily local) anatomical basis, which is why, for instance, there is such a consistent link between one fairly rare disorder (aneurysm of the anterior communicating artery) and the symptom. By contrast, however, the high frequency of planning problems indicated in this mixed aetiology group (for


213

group details see Burgess et al., 1998) argues for a complex informationprocessing basis and a correspondingly complex neuroanatomical substrate. If the root of these patients’ problems is one of “look-ahead”, then the processes which support that function are likely to be very diverse. In the context of the conclusion from the brief review above, it seems instead more plausible that planning is an activity which takes a number of forms and requires a great many quite different processes. So if situations that require planning (especially real-life ones) only minimally involve “look-ahead” processes, what might be the other processes involved?

PLANNING USING REAL-LIFE ANALOGUE TASKS Recently we have been working with a set of planning tasks that mimic reallife situations (Burgess, Coates, & Channon, in prep.). One of these tasks is outlined in Figure 10.1 and Box 10.1. In this planning test, subjects have to plan a shopping trip. They first learn a scenario, together with a set of things they have to achieve (e.g., get some medication for an infirm mother; feed the ducks at the local pond; send a birthday card), and a set of preferences or constraints they have to apply in choosing what they are going to do (e.g., you don’t enjoy shopping; you only have a watch and money with you; see Box 10.1). They are then free to form as good a plan as they are able, based on the information they have. Note that quite apart from the task requiring the

Figure 10.1 Display given to participants performing the Shopping Plan Test (SPT) (see Box 10.1 for test instructions, and Box 10.2 for an example test performance).

214


Box 10.1

Instructions (abbreviated) for the Shopping Plan test (SPT)

You are catching a bus to go and feed the ducks at the local village pond. You also need to do some shopping and sort out some other things: Things to do 1 Pick up a prescription and buy some medication for your infirm mother. 2 Buy a large dustbin. 3 Send a birthday card to a friend. 4 Find out the starting times of the feature films at the local cinema tonight. You need to bear in mind 1 The only items you have when you get off the bus are a watch and some money. 2 You can buy most of the things you need in the superstore (a large department store) but it always has very long queues and is no cheaper than the local shops. 3 You don’t like shopping very much and would like to spend as little time as possible doing it.

computation of an optimal strategy given the instructions, there are also many “hidden” matters to be discovered in a test like this which are not in the instructions, and therefore require further planning. For instance, what are you going to feed the ducks? How will you write the card given that you only have a watch and money on you? We have collected a series of verbal protocols of healthy people performing this task, describing their thought processes as they plan (Burgess et al., in prep.). A typical example is given in Box 10.2, which is what subject JS said whilst he was performing the shopping plan task outlined in Figure 10.1 and Box 10.1. It is immediately apparent that there are a large number of “control elements”, that in many ways mimic those found in protocols of autobiographical recollection (see Burgess & Shallice, 1996). These include many corrections and changes of mind (see e.g., points [2], [3], [5], [6], [7] and [11] in Box 10.2). It is clear that the planning proceeds by a series of preliminary hypothetical steps which are later firmed into a definite plan (see the plan summary given at the end of Box 10.2). However it is not the case that these subplans are particularly well thought through. For instance, in the protocol in Box 10.2 the subject makes a variety of decisions concerning the purchase of the dustbin, most of which are later overturned. It is by no means the case that he comes to a conclusion about what to do about the dustbin purchase when he first considers it (the most rational solution being to buy the dustbin in the hardware store on the way back to the bus stop), and then proceeds to the


Box 10.2

215

Subject JS: Planning a shopping trip

“Erm . . . I’m just, the first thing that’s just popped into my head is the nearest store to the bus stop [1], which is the hardware store, erm . . . I’ll go in there no I wouldn’t [2] I’d sort of go in there to buy the dustbin but then [3] I’m trying to minimize the amount of shops I’m going to go into, and I’d have to go separately into the pharmacy for drugs, and presumably not to come back on myself, go to the newsagent and post office, for the card and post the card, leave the dustbin to the superstore [4] and . . . but [5] at the same time I’m thinking that’s just as easy going to the hardware store at the beginning and so I just imagine myself walking down this road, and going to the hardware store first to buy a dustbin, except no [6] I’ll change my mind again, because the dustbins heavy I don’t want to be carrying it all along the road, the superstore’s right by the seat at the duckpond. But [7] then I’d have to walk back up again with it to get the bus home, but, so . . . go straight to the pharmacy to buy the medication, that’s the most important thing in my job [8]. Go round to the newsagents to buy a card, erm post it at the post office at the same time presuming [9] you could check out the cinema times there, either there or in the newsagents when you buy a card. And then feeding the ducks [10], then buying the dustbin on the way back to the bus stop. Erm no [11] maybe doing it before to get all the shopping out the way at once, buy the dustbin there. Sit down by the ducks, I realize I’m going in more shops than I’d like but I’m weighing it up with the queues [12], and the fact you’ve got to go into the pharmacy separately anyway and the doctor, both the newsagents and the post office are on the way. Might be easier just to jet off to one or the other on either side. That’s it.” Subject asked for a summary of the plan, and replies: “Erm first stop after the bus stop would be going straight to the pharmacy to pick up the prescription and buy the medication for my mum. Erm go round the corner, up to the newsagents to buy the card, chose a card for your friend and check out while I’m there the local cinema times. On the way round to the duckpond, go into the post office and post the letter and or checking cinema times there if there weren’t any in the newsagents. Erm passing the superstore and buying a dustbin from the superstore because it’d be too much trouble to carry a heavy dustbin all the way down this road. Erm and get all the shopping out the way and sit on the seat and feed the ducks.”

next steps, adding these in turn onto each other. Instead the order in which matters are considered is much more idiosyncratic, reminiscent of naturalistic studies of autobiographical recollection (see e.g., Burgess & Shallice, 1996). There is also evidence of quite fine judgement making (see points [3] and [12]), where a decision has to be made about two possible courses which

216


appear to be of equal or near equal suitability. There are what one might call “conditional subplans” (see point [9]), where someone decides that “If X, then I will do Y”, in other words that the decision about exactly what to do will be deferred until one is actually in the situation. Another instructive aspect is the evidence for the application of personal preferences and assignments of priority which go beyond the task instructions. For instance, concerning the dustbin purchase the subject remarks, “Erm no, maybe doing it [i.e., buying the dustbin] before, to get all the shopping out the way at once”. Apparently despite being aware that buying the dustbin on the way to the pond will result in having to carry a heavy dustbin all the way down the road on the way back, he is deciding at this point that his preference is to put up with this so that when he visits the ducks he will do so in the knowledge that all his shopping is complete. This decision remains in his final summary of his plan. This is not strictly “rational” perhaps, and is the application of a preference which goes beyond the information given in the instructions. One would very much doubt, for instance, that a simple computer program written to produce a plan for this task would come up with such a conclusion. Moreover there is evidence of the assignment of priority to the tasks. The instructions do not mention any item being more important than others, yet the subject says “go straight to the pharmacy to buy the medication, that’s the most important thing”. There is no reason to assume that getting the medication is any more or less important than, for example, sending the birthday card. There are many scenarios one might consider where the relative significance is quite the opposite (e.g., if the medication is being bought well in advance of when it might be needed). This appears to be a supposition based on experience and/or personal semantic knowledge, which again goes beyond the task instructions. In this way, it is easy to see how the processes involved in real-life planning situations might largely tap considerably different (and probably many more) processes than a task like the TOL test. There is indeed a look-ahead component to the shopping task outlined here. Witness JS’s attempts to minimize the number of stores he will have to enter (see point [4] in the protocol), which, prima facie, could well make similar demands to, for instance, attempting to solve a TOL problem in the given number of moves. However, there are also apparently spontaneously recalled elements which mimic those found in autobiographical recollection (see Burgess and Shallice, 1996 for examples of recollection protocols). So for instance at point [1] in the protocol, JS reports first thinking about what is the nearest shop to the bus stop, without any particular reason for this thought, and without it being the product of a reasoning process of which he was consciously aware. There is also the assignment of relative importance to tasks which go beyond the task instructions (e.g., witness JS’s decision to get all the shopping done before


217

feeding the ducks so he can relax, even if this does mean that he has to carry the shopping – including a heavy dustbin – all the way back down the street on his way home). We would argue that the constrained nature of problems like the Tower tests provides too structured a situation to exercise these sorts of cognitions. Yet they must be central to understanding how the brain supports the sorts of activities that one might label as “planning”.

EVIDENCE FROM NEUROLOGICAL PATIENTS The putative existence of such processes would receive considerable support if one could show patterns of impairments in neurological patients that would be predicted on theoretical grounds. There is now a small but clear cohort of supporting studies. It has been most clearly articulated perhaps by the outstanding studies by Vinod Goel and his colleagues (see above). However, most cases of investigations of higher level executive deficits also give examples which seem to lend support. For instance, Eslinger and Damasio (1985) report that their case EVR would, before going out for an evening meal, typically visit all the possible restaurants and consider in detail such factors as the seating arrangements, menu and so forth, but would still be unable to come to a decision about where to go. Prima facie this seems exactly the sort of symptom one would expect if the systems which enable a person to assign “weights” to different aspects of a scenario were damaged. Another consequence of damage to such systems might be inconsistent and peculiar assigning of priorities, and rather unconstrained solutions to problems. Consider Figure 10.2(a). This shows a typical performance of a healthy control subject on the Multiple Errands Test (MET) of Shallice and Burgess (1991). In this test, which takes place in a real-life street (see Alderman, Burgess, Knight, & Henman, 2003; Knight, Alderman, & Burgess, 2002 for other versions) subjects are given some money and an instruction sheet, and have to buy various items (e.g., a birthday card) and find out various bits of information (e.g., the exchange rate of the French franc) whilst following a set of arbitrary rules (e.g., don’t go into a shop more than once; spend as little time as possible). In this way the problem presented to the subjects is very similar to the one presented in the Shopping Plan Test (SPT) above. The difference is that with the MET they are not explicitly prompted to plan, and have to actually carry out the task in the real world rather than just say how it might be done. Now consider the performance of patient DN in Figure 10.1(b). DN’s performance not only contained a larger number of the sorts of errors that controls make (e.g., forgetting to get all the items in a shop that one might), but he also showed behaviour that was not shown by any control. For instance, when attempting to buy a birthday card (one of the items on his

Figure 10.2 (a) A schematic representation of a typical healthy control participant’s performance on the Shallice and Burgess (1991) Multiple Errands test, which is conducted in a shopping precinct. (b) The impaired performance of patient DN on this test (see text for patient details).


219

shopping list) he discovered that he had run out of money, so he suggested to the woman in the shop that he could exchange sexual favours for the card. (This strategy was not successful.) Perhaps the most significant difference, however, between DN’s performance and the controls was the apparently aimless wandering up and down the street. This suggested a lack of “on-line” planning (i.e., adapting the aim of one’s behaviour in response to unforeseen circumstances). Importantly, however, although DN was severely impaired in this everyday-type situation that requires planning, his performance on the TOL task was normal. Thus his performance may be an example of a deficit in the sorts of planning processes described above (i.e., other than “lookahead”).

PLANNING DEFICITS AND LOCALIZATION Let us then imagine that planning is an activity that can be supported by very many different cognitive processes – perhaps hundreds – and that different subsets of these are tapped by different planning tasks. One might expect two experimental consequences. The first is that tasks which have a strong planning component are likely to be quite sensitive to lesions in a wide range of brain regions, meaning that they will be quite clinically sensitive. The second is that small changes in task format might radically alter the construct demands of the task (i.e., the degree to which a task taps a specific putative process such as memory, inhibition, and so forth), and one would therefore perhaps expect low concordance between specific regions of brain damage and deficits on similar but different tasks. Alderman et al. (2003) give evidence of the first variety. This study looked at a simplified version of the Shallice and Burgess Multiple Errands Task (MET), and found that 82% of a mixed aetiology neurological patient group performed at or below the 5% level of the healthy controls. This degree of sensitivity is remarkable. Of course the MET requires processes other than planning (namely multitasking), but there is little doubt that “planning”, however it is envisaged, is a key component of the test. Evidence of the second type comes from recent studies in our laboratory. Burgess et al. (2000) were principally interested in the brain regions involved in multitasking, and administered a multitasking test to 60 patients with circumscribed cerebral lesions. This test was similar in format to the Six Elements Task (SET), except that there were fewer tasks to be attempted, and more rules to follow. After the participants had learnt the task instructions, and before they started doing the test, they were asked for their plan of what they intended to do. Burgess et al. found that people whose lesions involved the right dorsolateral prefrontal cortex produced significantly poorer plans than patients with lesions elsewhere (and healthy controls). Moreover, right

220


dorsolateral lesions did not cause impairments in other aspects of the task (e.g., learning and remembering task rules, recounting what one had done, etc.), so this deficit seems like good evidence for the localization of planning processes to this region. However, this finding was not replicated when the identical procedure was applied to a version of the Six Elements Task (SET) (Burgess et al., 2000). Moreover, when examining the cognitive profile of the cases with right dorsolateral prefrontal cortex (RDLPFC) lesions we found that they were also significantly poorer at the Raven’s Advanced Progressive Matrices, which is not generally thought of as a “planning task”. Thus, it is likely that the deficits in these patients are not in fact “planning specific”, and it is not the case that a right dorsolateral PFC lesion will necessarily mean that if that person is presented with a planning situation and asked “What do you intend to do?” they will produce a poor plan. Instead, the result would appear to depend upon the precise demands of the situation. (This will be the subject of further investigation.)

PLANNING AND “CONSTRUCT VALIDITY” Turning now to evidence from healthy people, as noted earlier in this chapter if there are processes which are invariably involved in planning, then one should in general expect higher correlations between people’s performances on different planning tasks than between planning tasks and non-planning ones. However, evidence from our laboratory suggests that this is not the case, even if the different planning tasks seem quite similar in format. Burgess et al. (in preparation) administered the shopping plan test (see Figure 10.1 and Box 10.1) and a similar task that required people to plan a party. The tests were similar in the amount of information that had to be learnt; the numbers of constraints and rules of the situation with which they were faced; and the format of presentation. However, in a group of 92 undergraduates there was no significant correlation between performance on the tasks, despite significant correlations between the planning tasks and verbal memory ones that had no obvious planning component. When the variance that could be attributable to the memory tasks was removed statistically, the correlations between the planning tasks reduced to close to zero. We were surprised by these findings and so administered the planning tasks to a separate group of 75 undergraduates, alongside a group of nonverbal memory tests (Burgess et al., in prep). The pattern of results was the same: no correlation between the different planning tasks despite significant correlations between the planning tasks and non-verbal memory ones. The precise psychometric conditions that may lead to these results will be the subject of some careful statistical modelling. However, some recent evidence from functional brain imaging studies of four of these tasks provides a possible explanation of them (Burgess, Scott, Frith, in press).


221

Using the H215O PET method we found regional cerebral blood flow (rCBF) increases in medial temporal lobe structures when people were planning (compared with non-planning control tasks that used the same stimulus materials) across all the tasks. These activations showed strong lateralization according to stimulus materials, with left hemispheric increases when the planning task was presented verbally, and right hemisphere where the task involved the presentation of a spatial display. We assume these activations to reflect the mnemonic demands of the task. Interestingly, however, we also found significant rCBF differences between the various planning tasks. It is unlikely that these differences reflect straightforward material-specific (e.g., spatial/verbal and so forth) differences between the tasks since each planning task had its own control, which was the identical display to that used in the planning conditions. Instead it seems likely that the task-specific rCBF differences reflect the subtly differing processing demands of the planning tasks. It is doubtful that functional imaging evidence could ever be used to dismiss categorically the possibility of common processing between tasks – the logic of the links between putative process and physiological changes does not permit such conclusions to be drawn. However, one can conclude at least that these PET results are not inconsistent with a view of the processes underlying planning as distributed across the brain, and apparently relatively small differences in task format having considerable consequences for the way the brain is required to work in order to perform them.

CONCLUSION There are at least three possible explanations for the results of our studies presented here, which have used analogues of real-world situations. The first possibility is the “no planning-specific processes” explanation. This would hold that there are no specific “planning processes”, and that “planning” is a behavioural label for a range of activities (like “travelling”) which share little in common in terms of their processing demands. However, performance of the tasks may share some more general “non-planning” processing requirements (e.g., memory) which explains the correlations with non-planning tasks. The second possibility is that the various planning tasks all tap “planning processes”, but different ones. A third is that different tasks tap similar planning processes, but the outcomes of that processing (i.e., the conclusions people reach about what they should do) differ from subject to subject since they are guided by personal preferences and conclusions from past experiences (just as in the example above where PB never checks the prices of foodstuffs). The first explanation, if true, would suggest that whilst it might be possible to achieve a behavioural-level explanation of the label “planning” (e.g., “a person is indulging in planning when they say X”), it would not be possible to achieve a satisfactory processing-level explanation (i.e., the activity called

222


“planning” necessarily requires information transformation of X type). This would present a challenge for cognitive neuroscientific enquiry in determining prototypical tasks; yet without them one has no kernel of enquiry with which to start. The experimental implications of the second explanation are actually quite similar in that it suggests that one would have to abandon the notion of a prototypical task (e.g., the role that the TOL task has been fulfilling for the last 20 years) and instead study a very large number of situations that call for the activity described as “planning”. The implication of the third explanation is that much of the experimentation in cognitive neuroscience to date is of limited use in understanding how the brain supports the activity people term “planning”. Instead, we should be studying how people determine what is important to them in a scenario, apply past experience in determining action and deciding how far ahead to look and so forth. There must be cognitive processes which support these functions but, as many authors have pointed out, they have been hardly studied at all (e.g., Cohen, 1996; Kafer & Hunter, 1997; Phillips, Wynn, Gilhooly, Della Sala, & Logie, 1999; Phillips, Wynn, McPherson, & Gilhooly, 2001; Ward & Allport, 1997). Interestingly, an account of this type might have the advantage of explaining other interesting phenomena reported in studies using real-world like activities. For instance, Garden, Phillips, and McPherson (2001) report little age-related performance decline in real-world situations requiring open-ended planning (versions of the Shallice & Burgess Six Elements and Multiple Errands tests). On the current account, this could be explained by the advantage older adults have from greater experience compensating for any decline in more fluid abilities such as “look-ahead”. Having outlined in broad terms three possibilities for the functional organization of planning processes, it would be nice to think that they could all be investigated independently in order to determine which best fits the data. However there is a serious constraint to this endeavour. If possibility (3) is true, then studies of possibilities (1) and (2) are unlikely to be misleading. This is because the influence of stored preferences upon performance will lead to apparent lack of association between tasks where, if the influence of stored preferences could be accounted for, there would in fact be strong association. If you remain unconvinced of the strength of influence of stored preferences in real-life situations, consider the following. In our data from the shopping task described above, one of the measures of performance we derived was whether people would work out that if one buys the dustbin on the way towards the duck pond, then one will be unnecessarily traipsing this heavy item both up and down the street. The test was therefore devised with a hardware store closest to the bus stop, so good planners would work out that the optimal plan was to buy the dustbin at the hardware store on the way home. However, our data indicated that this was actually a very unreliable indicator of performance. When we investigated the cause, we found that


223

women were significantly less likely to buy the dustbin at the hardware store than men, preferring to get it in the superstore instead (which is prima facie suboptimal for the reason already outlined, plus others). Canvassing a group of male and female subjects then revealed that women would often either deliberately avoid going to the hardware store because they are not used to going to such places, expecting them to be unfriendly, or just not really register its presence at all since they are not places with which they have experience. Even women whose plans were otherwise extremely good and who performed in the top 1 percentile on the other measures tended to show this pattern. Now, who is to say that the degree of real planning that has gone on in the conscious decision not to go to the hardware store is any less than would be involved in deciding to go into it? Thus in order to measure the construct of planning with this measure we may have to ignore the actual form of the final decision (e.g., hardware store = correct; superstore for bin = incorrect) and instead somehow measure the process of reaching it. This will require considerable experimental ingenuity. However, if we are to determine the organization of the cognitive system that supports planning, we must first understand the effects of processing which exert so much influence. Indeed, we have argued that the expression of stored preferences in a plan is planning. Turning to other experimental studies of planning, it is clear that the cognitive neuroscience of planning is currently dominated (at least in volume of output) by studies of the TOL and its variants. Our selective review of the cognitive neuroscience literature suggests however that it is by no means clear how much these studies have to say about how people create plans, or how the brain supports this activity. It seems likely that people with frontal lobe damage often (or perhaps always) fail Tower-type tasks not because of a “lookahead” deficit per se, but rather because of problems resolving the particular goal–subgoal conflict presented by certain test items, because of problems following rather than creating plans, or other difficulties. It may also be that task complexity plays some part. Indeed, it may even be the case that performance of Tower tasks only minimally requires “look-ahead”. This is a serious problem for cognitive neuroscience of planning. The study of these tasks is largely predicated on the view that they measure some processes that enable “look-ahead”. Without this assurance, the fact that the situation with which a subject is faced when asked to perform Tower-type tasks differs in very many ways from situations in everyday life that require planning (e.g., deciding what to do next weekend; planning a vacation or journey, etc.) becomes of great concern, since it is then by no means clear how the results are relevant to anything one might wish to know. We study experimental tasks as simplified “models of the world”. However, after 20 years of research into Tower-type tasks, we might have to accept that they are poor models of the world and do not actually measure what we originally thought they might.

224


The obvious solution is to create new “models of the world”. What form should these take? Perhaps the lesson that can be learnt from the Tower tasks is that it would be better to start by using tasks which very obviously mimic the requirements of everyday life planning situations. Whatever the subsequent results, at least then there would not be any argument about whether one was actually studying processes relevant to “planning”. However, the obstacle to this enterprise, as already noted, is that we actually know very little about how or why people form plans in everyday life, or the situations that provoke people to do so. In this laboratory, we have tried to start filling this gap by studying the analogues of everyday situations. The results suggest the three possibilities outlined above for the way that the brain supports planning. All three are to various degrees congruent with the conclusions of many authors who have studied Tower-type tasks, so the data provides little useful constraint for theorizing about the organization of the cognitive system that supports planning. In the absence of these constraints from traditional experimental studies, we currently favour starting with the third possibility above, since it would need to be discounted before progressing to the other possibilities. Moreover, although the third possibility is currently little better at explaining the quantitative aspects of the data we have described, it does concord better with the qualitative aspects (see the example in Box 10.2). For these reasons we are currently investigating the view that there may be some key processes which are characteristic of “planning”, but that the expression of them in terms of what someone decides to do is greatly determined by the influence of preferences and prejudices that one brings to situations. As we have said, it is our view that the application of these preferences is an integral part of the activity called “planning”, and that the study of how these preferences are formed and applied should therefore similarly be integral to research in this exciting area. Although this will present a considerable methodological and conceptual challenge, the results could be enormously rewarding. In any case, we may have no realistic alternative, since the investigative procedures most prevalent in cognitive neuroscience at the moment may have served their purpose.

REFERENCES Alderman, N., Burgess, P. W., Knight, C., & Henman, C. (2003). Ecological validity of a simplified version of the multiple errands shopping test. Journal of the International Neuropsychological Society, 9, 31–44. Baker, S. C., Rogers, R. D., Owen, A. M., Frith, C. D., Dolan, R. J., Frackowiak, R. S. J., & Robbins, T. W. (1996). Neural systems engaged by planning: A PET study of the Tower of London task. Neuropsychologia, 34, 515–526. Burgess, P. W., & Shallice, T. (1996). Confabulation and the control of recollection. Memory, 4, 359–411.


225

Burgess, P. W., & Robertson, I. H. (2002). Principles of the rehabilitation of frontal lobe executive deficits. In D. T. Dtuss & R. Knight (Eds.), Principles of frontal lobe function. New York: Oxford University Press. Burgess, P. W., Alderman, N., Emslie, H., Evans, J. J., & Wilson, B. A. (1996). The dysexecutive questionnaire. In B. A. Wilson, N. Alderman, P. W. Burgess, H. Emslie, & J. J. Evans (Eds.), Behavioural assessment of the dysexecutive syndrome. Bury St Edmunds: Thames Valley Test Company. Burgess, P. W., Alderman, N., Evans, J. R., Emslie, H., & Wilson, B. A. (1998). The ecological validity of tests of executive function. Journal of the International Neuropsychological Society, 4, 547–558. Burgess, P. W., Coates, L. M.-A., & Channon, S. (in prep.) Mechanisms of everyday planning. Burgess, P. W., Quayle, A., & Frith, C. D. (2001). Brain regions involved in prospective memory as determined by positron emission tomography. Neuropsychologia, 39, 545–555. Burgess, P. W., Veitch, E., de Lacy Costello, A., & Shallice, T. (2000). The cognitive and neuroanatomical correlates of multitasking. Neuropsychologia, 38, 848–863. Burgess, P. W., Scott, S. K., & Frith, C. D. (2003). The role of the rostral frontal cortex (Area 10) in prospective memory: A lateral vs. medial dissociation. Neuropsychologia, 41, 906–918. Butters, N., Wolfe, J., Martone, M., Granholm, E., & Cermak, L. S. (1985). Memory disorders associated with Huntington’s disease: Verbal recall, verbal recognition and procedural memory. Neuropsychologia, 23, 729–743. Carlin, D., Bonerba, J., Phipps, M., Alexander, G., Shapiro, M., & Grafman, J. (2000). Planning impairments in frontal lobe dementia and frontal lobe lesion patients. Neuropsychologia, 38, 655–665. Channon, S., & Crawford, S. (1999). Problem-solving in real-life-type situations: The effects of anterior and posterior lesions on performance. Neuropsychologia, 37, 757–770. Cohen, G. (1996). Memory in the real world (2nd ed.). Hove, UK: Psychology Press. Colvin, M. K., Dunbar, K., & Grafman, J. (2001). The effects of frontal lobe lesions on goal achievement in the Water Jug task. Journal of Cognitive Neuroscience, 13, 1129–1147. Dagher, A., Owen, A. M., Boecker, H., & Brooks, D. J. (1999). Mapping the network for planning: A correlational PET activation study with the Tower of London task. Brain, 122, 1973–1987. Dagher, A., Owen, A. M., Boecker, H., & Brooks, D. J. (2001). The role of the striatum and hippocampus in planning: A PET activation study in Parkinson’s disease. Brain, 124, 1020–1032. Dehaene, S., & Changeux, J. P. (1997). A hierarchical neuronal network for planning behavior. Proceedings of the National Academy of Sciences USA, 94, 13293–13298. Elliott, R., Sahakian, B. J., Michael, A., Paykel, E. S., & Dolan, R. J. (1998). Abnormal neural response to feedback on planning and guessing tasks in patients with unipolar depression. Psychological Medicine, 28, 559–571. Eslinger, P. J., & Damasio, A. R. (1985). Severe disturbance of higher cognition after bilateral frontal ablation: Patient EVR. Neurology, 35, 1731–1741. Fincham, J. M., Carter, C. S., van Veen, V., Stenger, V. A., & Anderson, J. R. (2002). Neural mechanisms of planning: A computational analysis using event-related fMRI. Proceedings of the National Academy of Sciences USA, 99, 3346–3351. Garden, S. E., Phillips, L. H., & McPherson, S. E. (2001). Midlife aging, open-ended planning and laboratory measures of executive function. Neuropsychology, 15, 472–482. Goel, V., & Grafman, J. (1995). Are the frontal lobes implicated in “planning” functions? Interpreting data from the Tower of Hanoi. Neuropsychologia, 33, 623–642. Goel, V., & Grafman, J. (2000). Role of the right prefrontal cortex in ill-structured planning. Cognitive Neuropsychology, 17, 415–436. Goel, V., Grafman, J., Tajik, J., Gana, S., & Danto, D. (1997). A study of the performance of patients with frontal lobe lesions in a financial planning task. Brain, 120, 1805–1822.

226


Goel, V., Pullara, S. D., & Grafman, J. (2001). A computational model of frontal lobe dysfunction: Working memory and the Tower of Hanoi task. Cognitive Science, 25, 287–313. Goldberg, T. E., Saint-Cyr, J. A., & Weinberger, D. R. (1990). Assessment of procedural learning and problem solving in schizophrenic patients by Tower of Hanoi type tasks. Journal of Neuropsychiatry, 2, 165–173. Grafman, J., Litvan, I., Massaquoi, S., & Stewart, M. (1992). Cognitive planning deficit in patients with cerebellar atrophy. Neurology, 42, 1493–1496. Hughes, C., Robbins, T. W., & Russell, J. (1994). Evidence for executive dysfunction in autism. Neuropsychologia, 32, 477–492. Kafer, K. L., & Hunter, M. (1997). On testing the validity of planning/problem-solving tasks in a normal population. Journal of the International Neuropsychological Society, 3, 108–119. Knight, C., Alderman, N., & Burgess, P. W. (2002). Development of a simplified version of the multiple errands test for use in hospital settings. Neuropsychological Rehabilitation, 12, 231–255. Koechlin, E., Basso, G., Pietrini, P., Panzer, S., & Grafman, J. (1999). The role of the anterior prefrontal cortex in human cognition. Nature, 399, 148–151. Koechlin, E., Corrado, G., Pietrini, P., & Grafman, J. (2000). Dissociating the role of the medial and lateral anterior prefrontal cortex in human planning. Proceedings of the National Academy of Sciences USA, 97, 7651–7656. Levin, H., Mendelsohn, D., Lilly, M., & Fletcher, J. (1994). Tower of London performance in relation to magnetic resonance imaging following closed head injury in children. Neuropsychology, 8, 171–179. McCarthy, R. A., & Warrington, E. K. (1990). Cognitive neuropsychology: A clinical introduction. London: Academic Press. Miotto, E. C., & Morris, R. G. (1998). Virtual planning in patients with frontal lobe lesions. Cortex, 34, 639–657. Morris, R. G., Downes, J. J., Sahakian, B. J., Evenden, J. L., Heald, A., & Robbins, T. W. (1988). Planning and spatial working memory in Parkinson’s disease. Journal of Neurology, Neurosurgery & Psychiatry, 51, 757–766. Morris, R. G., Ahmed, S., Syed, G. M., & Toone, B. K. (1993). Neural correlates of planning ability: Frontal lobe activation during the Tower of London test. Neuropsychologia, 31, 1367–1378. Morris, R. G., Rushe, T., Woodruff, P. W. R., & Murray, R. M. (1995). Problem solving in schizophrenia: A specific deficit in planning ability. Schizophrenia Research, 14, 235–246. Morris, R. G., Miotto, E. C., Feigenbaum, J. D., Bullock, P., & Polkey, C. E. (1997). The effect of goal–subgoal conflict on planning ability after frontal- and temporal-lobe lesions in humans. Neuropsychologia, 35, 1147–1157. Norman, D. A., & Shallice, T. (1986). Attention to action: Willed and automatic control of behavior. In R.J. Davidson, G. Schwartz, & D. Shapiro (Eds.), Consciousness and self-regulation. New York: Plenum Press. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1990). Planning and spatial working memory after frontal lobe lesions in man. Neuropsychologia, 28, 1021–1034. Owen, A. M., James, M., Leigh, P. N., Summers, B. A., Marsden, C. D., Quinn, N. P., & et al. (1992). Fronto-striatal cognitive deficits at different stages of Parkinson’s disease. Brain, 115, 1727–1751. Owen, A. M., Doyon, J., Petrides, M., & Evans, A. C. (1996). Planning and spatial working memory: A positron emission tomography study in humans. European Journal of Neuroscience, 8, 353–364. Owen, A. M., Doyon, J., Dagher, A., Sadikot, A., & Evans, A. C. (1998). Abnormal basal ganglia outflow in Parkinson’s disease identified with PET: Implications for higher cortical functions. Brain, 121, 949–965.


227

Phillips, L. H., Wynn, V. E., Gilhooly, K. J., Della Sala, S., & Logie, R. H. (1999). The role of memory in the Tower of London task. Memory, 7, 209–231. Phillips, L. H., Wynn, V. E., McPherson, S. E., & Gilhooly, K. J. (2001). Mental planning and the Tower of London task. Quarterly Journal of Experimental Psychology, 54, 579–597. Rowe, J. B., Owen, A. M., Johnsrude, I. S., & Passingham, R. E. (2001). Imaging the components of a planning task. Neuropsychologia, 39, 315–327. Shallice, T. (1982). Specific impairments in planning. Philosophical Transactions of the Royal Society of London, B298, 199–209. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T., & Burgess, P. W. (1991). Deficits in strategy application following frontal lobe damage in man. Brain, 114, 727–741. Simon, H. A. (1975). The functional equivalence of problem solving skills. Cognitive Psychology, 7, 268–288. Simons, J. S., Verfaellie, M., Galton, C. J., Miller, B. L., Hodges, J. R., & Graham, K. S. (2002). Recollection-based memory in frontotemporal dementia: Implications for theories of long-term memory. Brain, 125, 2523–2536. Veale, D. M., Sahakian, B. J., Owen, A. M., & Marks, I. M. (1996). Specific cognitive deficits in tests sensitive to frontal lobe dysfunction in obsessive-compulsive disorder. Psychological Medicine, 26, 1261–1269. Ward, G., & Allport, D. A. (1997). Planning and problem-solving using the 5-disk Tower of London task. Quarterly Journal of Experimental Psychology, 50, 49–78. Xu, Y., & Corkin, S. (2001). H.M. revisits the Tower of Hanoi puzzle. Neuropsychology, 15, 69–79. Zalla, T., Plassiart, C., Pillon, B., Grafman, J., & Sirigu, A. (2001). Action planning in a virtual context after prefrontal cortex damage. Neuropsychologia, 39, 759–770.

Author index

Abelson, R.P., 16–17, 62, 182–183 Adelson, B., 64 Adey, W.R., 145 Agid, Y., 190 Ahlum-Heath, M.E., 42 Ahmed, S., 35, 79, 82, 136, 142, 189, 206 Ajzen, I., 23 Alba, J.W., 57 Alderman, N., 120, 124–125, 128, 166, 211–213, 217, 219 Alexander, G., 205 Alexander, M.P., 120 Alivisatos, B., 146 Allen, J.F., 181 Allinger, R., 142 Allmano, N., 81, 113 Allport, D.A., 40, 79, 90, 92, 94–99, 105–106, 188, 222 Anderson, E.J., 21 Anderson, J.R., 3, 10, 19–20, 35, 42, 53, 55–57, 68, 76, 96, 139, 183–184, 207 Anderson, S.W., 190–191 Anderson, V., 116 Andrade, J., 21, 71 Andreasen, N.C., 142

Andrés, A.U., 111, 113–114, 125 Andres, P., 27 Anzai, Y., 7, 44–45, 79, 96 Arikuni, T., 145 Armitage, C.J., 23 Atkinson, R.C., 76 Atwood, M.E., 64, 79 Awh, E., 146 Baddeley, A.D., 20–22, 28, 76–77, 79, 81–82, 84, 90, 92, 115 Bajwa, A., 140 Baker, R., 189 Baker, S.C., 113, 136, 143, 145, 189, 206 Baker-Sennett, J., 183 Ball, L.J., 62, 65–66 Ballard, D.H., 140 Ballas, J.A., 96 Barbas, H., 145 Barch, D.M., 94, 96 Barker, D.R., 21 Barker, R.A., 143 Barnes, C.L., 142 Basso, G., 208 Bearman, C., 56 Bechara, A., 190–191

229

230

AUTHOR INDEX

Bensinge, D.G., 140 Benson, D.F., 26, 35 Benson, J.B., 194 Berardi-Coletta, B., 42 Berger, J., 94 Berman, K.F., 146 Bernard, S., 130, 153, 166, 189 Berns, G.S., 191 Berry, D.C., 42 Bianchi, L., 26, 135 Biederman, I., 93–94 Bisiacchi, P.S., 115, 124, 126 Blamire, A.M., 146 Bloch, G., 146 Blum, A., 184 Boecker, H., 136, 206 Bonerba, J., 205 Botvinik, M.M., 94, 96 Bradley, A.C., 21, 84 Bramham, J., 25, 27, 29–30, 147, 153–179, 185 Brandimonte, M.A., 79 Brandt, S.A., 141 Braver, T.S., 94, 96, 192 Brennan, M., 79 Brereton, N., 81 Broadbent, D.E., 76 Brodmann, K., 144–145 Brooks, B., 153, 171, 173, 177 Brooks, D.J., 136, 143, 145–146, 206 Brown, D.L., 146 Bryan, J., 113, 131 Bullock, P., 35, 146, 153, 156, 158–163, 165, 171, 189, 205–206 Burack, O.R., 124 Burgess, P.W., 25–27, 29, 31, 92, 116, 119–120, 122, 124–125, 128, 130, 153, 155, 161, 166, 175, 189, 199–227 Burton, A.M., 42 Butters, N., 205 Buyer, L.S., 42 Byrne, R.W., 37

Call, J., 35 Callender, J.S., 124–125

Card, S., 43 Carl, J.R., 146 Carlin, D., 205 Carpenter, P.A., 78, 194 Carter, C.S., 35, 94, 96, 207 Castell, A.M., 42 Cermak, L.S., 205 Chalmers, D., 124, 129 Changeux, J.P., 139, 188, 207 Channon, S., 25–26, 29, 31, 199–227 Chapman, D., 184 Charness, N., 83–84 Chase, W.G., 61, 83 Chater, N., 57 Christal, R.E., 78 Chronicle, E.P., 57–60, 68 Cicerello, A., 79 Clark, A., 141 Clark, K., 181, 191, 195 Coates, L.M.-A., 25–26, 29, 31, 199–227 Cockburn, J., 168 Cocking, R.R., 4, 23, 184 Cohen, G., 71, 142, 222 Cohen, J.D., 93–96, 192 Colvin, M.K., 206 Conner, M., 23 Connor, L.T., 114 Conway, A.R.A., 78 Cools, R., 143 Cooper, V., 188 Corkin, S., 205 Corrado, G., 208 Corsi, A.T., 81–83 Courtney, S.M., 146, 192 Crawford, J.R., 113, 124–125, 127, 131 Crawford, S., 210 Cuneo, R., 79

Dagher, A., 135–136, 143, 145–146, 148, 205–206 Damasio, A.R., 153, 155, 190–191, 217 Damasio, H., 190–191 Daneman, M., 78 Danto, D., 25, 166, 190, 207, 209 Das, J.P., 183

AUTHOR INDEX

Davies, S.P., 10, 12, 17, 25, 27–28, 35–51, 53, 55, 60, 63–64, 68, 72, 107 De Jong, R., 95 de Lacy Costello, A., 208, 219–220 DeGroot, A.D., 61 Dehaene, S., 139, 188, 207 Dehn, D., 77 Del Missier, F., 42 Delaney, P.F., 84 Della Sala, S., 21, 39, 45–46, 76, 79–83, 106, 111, 113–115, 124, 125, 128, 130, 222 Desimone, R., 192 D’Esposito, M., 27 Deuschl, G., 191 Di Vesta, F.J., 42 Dolan, R.J., 113, 136, 143, 145, 189, 206 Dominowski, R.L., 42 Downes, J.D., 136–138, 145–146 Downes, J.J., 35, 135–136, 143, 156, 158, 185, 191, 205 Doyon, J., 79, 82–83, 113, 135–136, 142, 143–146, 185, 205–206 Dreher, M., 183 Driver, J., 91–92 Drummond, M., 181 Dubois, B., 190 Dunbar, K., 93–95, 206 Duncker, K., 54–56, 72 Dunlosky, J., 114

Egan, D.E., 159 Einstein, G.O., 121–122, 129 Elliott, R., 189, 205 Ellis, J.A., 121 Ellis, S., 107 Emerson, M.J., 85 Emslie, H., 120, 124–125, 128, 166, 211–213 Engle, R.W., 78 Englekamp, J., 77 Erhlich, K., 61, 68 Erickson, C.A., 192 Ericsson, K.A., 43, 75, 80, 84 Eslinger, P.J., 153, 155, 217

231

Evans, A.C., 79, 82–83, 113, 135–136, 142–146, 185, 205–206 Evans, J.J., 120, 124–125, 128, 154–155, 166, 211–213 Evenden, J.L., 135–136, 143, 156, 158, 191, 205 Ewert, O., 111, 118–119, 126, 128–130 Eysenck, M.W., 68 Farinello, C., 115, 124, 126 Fearnyhough, C., 21 Feigenbaum, J.D., 35, 114, 146, 159–163, 165, 189, 205–206 Fenwick, P.B.C., 130, 153, 166, 189 Fikes, R.E., 38, 183 Fincham, J.M., 35, 207 Fletcher, J., 205 Flitman, S., 188, 191 Forshaw, M.J., 80, 82, 115, 128 Frackowiak, R.S.J., 113, 136, 143, 145, 189, 206 Frensch, P.A., 42 Friedman, N.P., 85 Friedman, S.L., 2, 4, 23–24, 107, 181–184 Frith, C.D., 113, 136, 143, 145, 189, 206, 208, 220 Fritz, C.O., 66, 68 Fum, D., 42 Furneaux, S., 140 Furst, M., 184 Fuster, J.M., 27, 192 Gagne, R.M., 42 Galanter, E., 3, 12–17, 20, 22, 71, 75, 91, 181–183 Galton, C.J., 205 Gana, S., 25, 166, 190, 207, 209 Garavan, H., 118 Garden, S.E., 119–120, 124, 125, 222 Garnham, A., 159 Gazzaniga, M.S., 135 Gentner, D., 56 Gerbino, W., 79 Gick, M.L., 55–56 Gilhooly, K.J., 10, 12, 17, 21, 25, 27–30,

232

AUTHOR INDEX

39–40, 45–46, 71–88, 106, 111, 113–115, 125, 128, 130, 222 Gilmore, D.J., 44 Glosser, G., 82 Goel, V., 25, 35, 158, 166, 189–190, 205–207, 209–210, 217 Gold, J.M., 146 Goldberg, T.E., 146, 205 Goldman-Rakic, P.S., 145–146, 192 Goldstein, L.H., 130, 153, 166, 189 Goodglass, H., 82 Grafman, J., 25–27, 29, 31, 35, 38, 45, 68–69, 145, 158, 166, 181–198, 200, 205–211 Graham, K.S., 205 Granholm, E., 205 Gray, C., 81 Greeno, J.G., 159 Greenwood, P.M., 130 Groome, D., 68 Grossman, M., 27 Guindon, R., 65 Gur, R.C., 111 Gur, R.E., 111

Haider, H., 42 Haith, M.M., 194 Hallett, M., 191 Halsband, U., 42, 106 Hamilton, S., 124, 125 Hammond, K., 181 Harlow, J.M., 135, 154 Harnishfeger, K.K., 129 Harward H., 68–69 Hasher, L., 118, 129 Hawkins, J., 183 Haxby, J.V., 146, 192 Hayes, J.R., 7, 10, 41, 53 Hayes-Roth, B., 3, 11–12, 24, 37, 65, 115, 182–183 Hayes-Roth, F., 3, 11–12, 24, 37, 65, 115, 182–183 Haygood, R.C., 44 Hayhoe, M.H., 140 Heald, A., 135–136, 143, 156, 158, 191, 205

Hebb, D.O., 26 Hegarty, M., 44 Hendler, J., 182 Henik, A., 94 Henman, C., 217, 219 Henry, J.D., 127 Henson, R., 21 Hertzog, C., 114 Hitch, G.J., 20–21, 76–77, 79, 115 Hoare, C.A.R., 63 Hoc, J.-M., 10–11, 181 Hodges, J.R., 135–136, 138, 140, 143, 145–146, 189, 205 Hodgson, T.L., 136, 140, 148 Holding, D.H., 46, 62, 83–84 Holyoak, K.J., 55–56 Hommel, B., 92 Howard, H., 35, 38, 45 Howerter, A., 85 Hseih, S., 94–95 Hudson, S.R., 21, 84 Hughes, C., 35, 205 Hunt, E.B., 44 Hunter, M., 222 Huston, T.A., 94 Hyder, F., 146 James, M., 113–114, 125, 131, 135–136, 143, 146, 205 Jeannerod, M., 92 Jeffries, R., 42, 64, 79 Jersild, A.T., 93 Johnson, D.F., 44 Johnson-Laird, P.N., 23 Johnsrude, I.S., 136, 146, 206–207 Jones-Gotman, M., 158 Jonides, J., 146, 192 Jouandet, M., 135 Joyce, E.M., 79 Just, M.A., 78, 194 Kafer, K.L., 222 Kane, M.J., 78 Kar, B.C., 183 Karat, J., 39, 96 Karnath, H.O., 188, 191

AUTHOR INDEX

Karpov, B.A., 156 Kautz, H.A., 181 Keane, M.T., 68 Keil, K., 146, 192 Kennard, C., 136, 140, 189 Kieras, D.E., 96 King, J., 78 Kintsch, W., 41, 84 Kleist, K., 26 Kliegel, M., 10, 21, 25, 29, 40, 79, 111–133, 166 Knight, C., 217, 219 Koechlin, E., 208 Koeppe, R.A., 146 Koh, K., 56 Kolodner, J., 182 Kotitsa, M., 25, 27, 29–30, 147, 153–179, 185 Kotovsky, K., 7, 10, 41 Kray, J., 111 Kubota, K., 145 Kwee, S., 146 Kyllonen, P.C., 78

Lachman, M.E., 124 Laiacona, M., 113 Laird, J.E., 43, 183 Laming, D.R.J., 94 Land, M.F., 140 Lang, S., 65 Langenecker, S.A., 118 Langley, P., 181 Larkin, J.H., 41, 46 Lauber, E.J., 96 Laughlin, J.E., 78 Lawrence, A.D., 113–114, 125, 131 Lawrence, J.A., 124, 129 Lebiere, C., 21–22 Lee, K.Y., 191, 195 Lehto, J., 82 Leigh, P.N., 135–136, 143, 205 Leonhart, R., 42, 106 Levelt, W.J.M., 43 Levin, H., 35, 38, 45, 68–69, 205 Levine, B., 120 Lewis, R.L., 22

233

Lezak, M.D., 82 Lilly, M., 205 Lindenberger, U., 111 Lingard, A.R., 192 Litvan, I., 191, 205 Logie, R.H., 21–22, 39, 45–46, 76–77, 79–83, 106, 111, 113–115, 125, 128, 130, 222 Lou, J.-L., 181, 191 Lovett, M., 21–22 Lowenstein, J., 56 Lucking, C.H., 191 Luria, A.R., 26, 153, 156 Luszcz, M.A., 113, 131

McCarthy, G., 146 McCarthy, R.A., 201, 204 McClelland, J.L., 93–95 McDaniel, M.A., 121–122, 124–126, 129–130 McDermott, J., 46 MacDonald, R., 120 McGeorge, P., 42, 124, 125 MacGregor, J.N., 57–60, 68 McInnes, L., 113–114, 125, 131, 146 MacLeod, C.M., 44, 94 MacLeod, M.S., 10, 21, 25, 29, 40, 79, 111–133, 166 McNeil, J., 130, 153, 166, 189 McPherson, S.E., 21, 39–40, 73, 106, 111, 113, 119–120, 123, 125, 130, 222 Mannes, S.M., 41 Marks, I.M., 205 Marsden, C.D., 135, 143, 205 Martin, M., 111, 118–119, 121–122, 124, 126, 128–130 Martone, M., 205 Massaquoi, S., 181, 191, 205 Mathews, N.N., 44 Matusov, E., 183 Mayer, N.H., 185 Mendelsohn, D., 205 Metcalfe, J., 57 Meyer, D.E., 96 Meyer, E., 146 Meyer, M., 145

234

AUTHOR INDEX

Michael, A., 205 Middleton, F.A., 145 Milberg, W.P., 120 Miletich, R., 191, 195 Miller, B.L., 205 Miller, E.K., 192 Miller, G.A., 3, 12–17, 20, 22, 71, 75–76, 91, 181–183 Milne, A.B., 124, 125 Milner, B., 35, 146, 158, 165 Minoshima, S., 146 Minsky, M., 62 Mintun, M.A., 146 Minutun, M.A., 146 Miotto, E.C., 35, 114, 116, 129, 146, 156, 158–163, 165–166, 189, 205–206, 209 Miyake, A., 20, 22, 71, 82, 85 Monsell, S.M., 91–95 Montgomery, M., 185 Morris, R.G., 1–35, 37, 79, 82, 89, 91, 114, 116, 129, 135–136, 142–143, 146, 153–179, 185, 188–189, 191, 205–206, 209 Mueller, S., 96 Murray, R.M., 205 Nauta, W.J.H., 145 Neisser, U., 62 Newell, A., 3–4, 7–10, 15, 17–19, 22, 35–37, 43–44, 64, 75–76, 89–92, 96, 139, 182–183 Newman, S.D., 194 Nichelli, P., 191, 195 Nielson, K.A., 118 Nilsson, N.J., 38, 68, 183 Nimmo-Smith, I., 81, 118, 129 Nobre, A.C., 146 Noll, D.C., 192 Norman, D.A., 22, 26, 90–92, 116, 169 Nystrom, L.E., 192 Oakhill, J., 159 Oaksford, M., 57 Oatley, K., 23

Obonsawin, M.C., 113, 131 Obrist, W.D., 111 Oerter, R., 183 O’Hara, K.P., 41, 106 O’Leary, D.A., 189 O’Leary, D.S., 142 Ormerod, T.C., 10, 17, 25, 27–28, 53–70, 72 Owen, A.M., 25, 27, 29–30, 35, 71, 79, 82–83, 113–114, 125, 131, 135–151, 154, 156, 158, 181, 185, 205–207

Palmer, C., 185 Pandya, D.N., 142, 144–145 Panzer, S., 208 Papagno, C., 191 Parrila, R.K., 183 Parslow, D., 164 Partiot, A., 191 Pascual-Leone, A., 181, 191 Pasetti, C., 113 Passingham, R.E., 136, 146, 206–207 Patalano, A.L., 37 Pavlov, I.P., 26 Paykel, E.S., 189, 205 Payne, S.J., 41, 106 Pea, R.D., 183 Pelavin, R.N., 181 Penfield, W., 154–155 Pennington, B.F., 194 Pentland, L., 116 Perlstein, W.M., 192 Petrides, M., 35, 79, 82–83, 113, 136, 142, 144–146, 158, 165, 185, 206 Phillips, L.H., 10, 21, 25, 29, 39–40, 45–46, 73, 76, 79–83, 95, 106, 111–133, 166, 222 Phipps, M., 205 Pietrini, P., 191, 195, 208 Pillon, B., 190, 209 Plassiart, C., 209 Polkey, C.E., 35, 135–138, 140, 143, 145–146, 153, 156, 158–163, 165, 171, 185, 189, 205–206 Polson, P.G., 64, 79

AUTHOR INDEX

Pribram, K.H., 3, 12–17, 20, 22, 26, 71, 75, 91, 181–183 Puce, A., 146 Pullara, S.D., 35, 207 Quayle, A., 208 Quinn, N.P., 135, 143, 205 Rabbitt, P.M.A., 113–114, 125, 131, 146 Rahm, B., 42, 106 Randolph, C., 146 Rattermann, M.J., 29–30, 35, 38, 45, 68–69, 145, 181–198 Rau, P.S., 47 Raven, J.C., 81 Raz, N., 111 Razran, L., 79 Reason, J.T., 92 Reder, L.M., 21–22 Reed, S.K., 53, 57, 67, 79 Reed, E.S., 185 Reisberg, D., 68 Reitman, W.R., 72 Rellinger, E.R., 42 Reynolds, J.R., 46, 62 Rezai, K., 142 Richards, E.B., 192 Ridgeway, V., 118, 129 Ridgway, J., 66, 68 Rist, R.S., 46, 61–62, 68 Rivich, M., 111 Robbins, T.W., 21, 35, 79, 84, 113–114, 125, 131, 135–138, 140, 143, 145–146, 156, 158, 185, 189, 191, 205–206 Roberts, M.J., 44, 95 Roberts, R.J., Jr., 194 Robertson, I.H., 118, 129, 211–212 Rodderinkhof, K.R., 92 Rogers, R.D., 91, 94–95, 113, 136, 143, 145, 189, 206 Rogoff, B., 183 Rose, D., 153, 171 Rose, F.E., 171, 173, 177 Rosenberg, D.R., 146 Rosenbloom, P.S., 44, 183 Rowe, J.B., 136, 146, 148, 206–208

235

Rudkin, S., 77 Ruff, C.C., 42, 106 Rushe, T., 205 Russell, J., 35, 205

Saariluoma, P., 84 Sacerdoti, E.D., 3, 10–11, 182–183 Sadato, N., 191 Sadikot, A., 205 Sahakian, B.J., 35, 113–114, 125, 131, 135–138, 140, 143, 145–146, 156, 158, 185, 189, 191, 205 Saint-Cyr, J.A., 205 Satterlee-Cartmell, T., 82, 113 Schank, R.C., 16–17, 62, 182–183 Scholnick, E.K., 2, 4, 23–24, 107, 181–184 Schumacher, E.H., 146 Schwartz, M.F., 120, 185 Schwartz, M.L., 145 Scott, S.K., 208, 220 Sebrechts, M.M., 47 Seifert, C.M., 37 Sejnowski, T.J., 191 Selemon, L.D., 145 Semple, J., 138 Seymour, T., 96 Sgaramella, T.M., 115, 124, 126 Shah, P., 20, 22, 71, 82 Shallice, T., 4–5, 22, 26–27, 35, 82, 90–93, 116, 119, 122, 124–125, 130, 135–136, 155, 158, 161, 165–166, 169–170, 175, 189, 200, 202–205, 208–210, 212, 214–220, 222 Shapiro, M., 205 Shaw, J.C., 3–4, 8, 15, 17, 37, 90 Shiffrin, R.M., 76 Shulman, R.G., 146 Shute, V.J., 78 Siegler, R.S., 107 Simon, D., 46 Simon, H.A., 3–4, 7–10, 15, 17–19, 22, 35–37, 41, 43–46, 53, 57, 61, 63–64, 67, 75, 79–80, 83, 89–91, 96, 139, 159, 165, 182, 204 Simons, J.S., 25–26, 29, 31, 199–227

236

AUTHOR INDEX

Simplicio-Filho, F., 37, 43 Sims, V.K., 44 Sirigu, A., 190–191, 209 Skolnick, B.E., 111 Smith, E.C., 42 Smith, E.E., 146 Smith, L., 114 Smith, M.L., 35 Soloway, E., 61–62, 64, 68 Solso, R.L., 68 Spector, A., 93–94 Spector, L., 29–30, 35, 38, 45, 68–69, 145, 181–198 Spinnler, H., 113 Stark, L.W., 141 Stefanova, E., 143 Stenger, V.A., 35, 207 Sternberg, R.H., 44 Stewart, L., 113, 131 Stewart, M., 181, 191, 205 Stine, M., 82, 113 Strick, P.L., 145 Stroop, J.R., 93–96, 119 Stuss, D.T., 26, 120, 202–203 Styles, E.A., 94–95 Suchman, L.A., 181 Summers, B.A., 135–136, 138, 140, 143, 145–146, 205 Sussman, G.J., 11 Swayze, V., 142 Sweeney, J.A., 146 Syed, G.M., 35, 79, 82, 136, 142, 189, 206

Tajik, J., 25, 166, 190, 207, 209 Tennenberg, J.D., 181 Theeuwes, J., 92 Thomas, J.C., Jr., 79 Thompson, L., 56 Tiesman, B., 136 Todd, J.A., 116 Tomasello, M., 35, 79 Toone, B.K., 35, 82, 136, 142, 189, 206 Tuholski, S.W., 78 Turner, A.A., 64

Tzelgov, J., 94 Ungerleider, L.G., 146, 192 Unterrainer, J.M., 42, 106 Van der Linden, M., 111, 113–114, 125 van Veen, V., 35, 207 VanLehn, K., 54 Varma, S., 194 Veale, D.M., 205 Veitch, E., 208, 219–220 Verfaellie, M., 205 Visser, W., 65 Wachs, J., 191 Wagner, T.D., 85 Waldinger, R., 38 Walker, P., 79 Wallesch, C.W., 188, 191 Wallner-Allen, K.E., 24 Ward, G., 1–34, 37, 40, 79, 89–110, 136, 158–159, 185, 188, 222 Ward, T., 118, 129 Warrington, E.K., 201 Watanabe-Sawaguchi, K., 145 Wechsler, D., 81 Weibe, D., 57 Weinberger, D.R., 146, 205 Weisberg, R.W., 57 Welsh, M.C., 79, 82, 113 West, R.L., 111, 130 Wild, K., 191 Wilensky, R., 36, 182 Wilson, B.A., 120, 124–125, 128, 166, 211–213 Wilson, L., 81 Wiseman, M.B., 146 Witzki, A.H., 85 Wolfe, J., 205 Wood, J.N., 192 Woodruff, P.W.R., 205 Woods, D.J., 44 Wynn, V.E., 21, 39–40, 45–46, 73, 76, 79–83, 106, 111, 113–115, 125, 128, 130, 222

AUTHOR INDEX

Xu, Y., 205 Yarbuss, A.A., 156 Young, R.M., 22

Zacks, R.T., 118, 129 Zalla, T., 190, 209 Zhang, J., 41 Zimmermann, P., 188, 191

237

Subject index

ABSTRIPS problem solver, 10–11 ACT* model, 76 ACT-R architecture, 10, 19–20 working memory in, 21–22 Action plans control of, 93–96 lower level, 92–93 Action sets, 182 Actions goal-directed, 23–24 planning and executive control of, 89–110 Age–speed factor, 82 Ageing and action planning, 119–122 cognitive, 26 and cognitive planning, 111–133 and errand planning, 122–124 Amnesia medial temporal lobe, 205 Amygdalo-hippocampectomy, 138 Aneurysm, 212 Anterior cingulate cortex, 96 Arithmetic span, 78 Artificial Intelligence (AI), 10–11, 26, 35, 38, 181–182, 204

models of planning, 45, 68, 73 working memory capacity in, 76 Artificial planning task, 124–125 Attention measures of, 118, 129 Autism, 35, 205 Basal ganglia, 26, 31, 194 damage to, 130, 135 pathology of, 143, 185, 191 Behaviour dysregulation of, 26 and effort of will, 14 human, 91 opportunistic, 37 organization and structure of, 16–17 role of frontal lobes in, 26 structure of, 12–15 Behavioural Assessment of Dysexecutive Syndrome (BADS) tasks, 120, 128, 166 Best first approach, 74 Blackboard, 2, 12 Blocks world, 18–19 Blue ball problems, 141

239

240

SUBJECT INDEX

Brain hemispheric involvement in tasks, 79 lesions, 30, 79 mapping planning processes to, 193 Brain damage, 26 focal, 27, 153–179, 205 Breadth-first strategy, 63–66, 73–74 Broca’s area, 153 Brodmann areas, 145, 189 Bungalow task, 171–175 strategy formation in, 175 C language, 65 Cases, 182 Caudate nucleus, 143, 145 Central executive, 20, 76, 90 Cerebellar atrophy, 191, 205 Cerebellum pathology of, 185, 191 Cerebral blood flow (CBF), 142–143 145 regional (rCBF), 221 Chess, 9, 15, 21 blindfold, 84 large problem space of, 60–61 and perceptual recognition, 61 planning in, 83–84 process models in, 75–76 semantically rich problems of, 72–73, 83 working memory in, 83–84 Children-first strategy, 63–66 Cingulate cortex in route finding, 188 Closed head injury, 205 Cognition general theories of, 17–20 human, 3 Cognitive branching, 208 Cognitive functions measured by planning tasks, 127–128 Cognitive neuroscience perspectives on planning, 185–191 Cognitive planning human, 135–151

Cognitive planning research, 23–24 relative absence of, 68 Cognitive processes higher order, 89 Cognitive psychology planning overlooked in, 68 Cognitive science, 181 Computer simulations, 7 Condition–action statements, 7 Confabulation, 212 Connectionist networks, 93–94 Construct validity, 200, 203 Contention scheduling, 22, 26, 90, 169, 171 Contextual planning task, 124–125 Controlled attention, 78 Cooking task, 99–100, 104 Corpus callosum, 96 Corsi block span, 81 Corsi block task, 82–83 Corsi distance estimation task, 82–83 Crisis goals, 16–17 Criterion failure, 58 Cryptarithmetic process models in, 75–76 puzzles, 9 Decomposition control strategies for, 63–66 Deconstruction, 92 Delayed response test, 119 Dementia fronto-temporal, 205 Depression, 205 Depth-first strategy, 63–66, 73 Design Fluency task, 158 DEX questionnaire, 211 Digit span, 81 Digit Stroop costs, 95 Dorsolateral cortex (DLPFC), 153, 206, 208, 211, 220 Dual-task approach, 81, 83–84 Dynamic memory, 18 Dysexecutive syndrome, 93, 211–212

SUBJECT INDEX

Ecological validity, 25, 31, 177 Educational task design, 66–67 Eight-coin problem, 60 8-Puzzle, 106 English as a second language (ESL) task design in, 66–67 Episodic buffer, 76–77 Episodic memory, 200 Errand planning tasks, 25, 115–116, 124, 128 and age, 122–124 as measure of plan formulation, 127 EVR, case of, 155, 217 Executive control, 2, 17 and planning, 22–23, 29 of thought and action, 89–110 Executive function age and frontal lobes, 111–112 role in planning, 128–129 Executive Golf task, 156–158 Executive Process/Interactive-Control (EPIC) model, 96 Expertise as plan recall, 61–63 Exploratory factor analysis, 82 Eye movements, 140 after brain damage, 156 role in task activity, 30 Eye-tracking behaviour and cognitive performance, 139–142 Financial Planning task, 25, 166, 190, 209 Fluid intelligence, 78, 81 Fortress problem, 56 Frames, 62 Frontal cortex, 135, 142 activation in, 83, 147 areas of, 144–145 cerebral blood flow changes in, 143 in monkeys, 142, 145 Frontal gyrus, 208 Frontal lobes, 26–27, 35, 96, 135 activation of, 83 age and executive function, 111–112

241

in age-related planning impairment, 130–131 areas of, 153 damage/injury to, 29, 93, 114, 119, 122, 138–139 effects of surgery on, 154–155 and equivocation, 163–165 lesions of, 31, 189–191, 209 Frontopolar cortex, 208 Functional magnetic resonance imaging (fMRI), 142, 144, 148, 221 Gage, Phineas, case of, 154 Gambling task, 190 General Problem Solver program, 9–10, 36 Gestalt, 57 Goal-selection strategy, 80 Goal–subgoal conflict, 114, 159–161, 206–207, 211, 223 HACKER model, 11 Haemodynamic response functions, 207 Heuristics, 2–3, 17 hill climbing, 5, 9, 53, 74, 79, 159, 165 inapplicability to ill-defined problems, 54 means-ends analysis, 5, 9–10, 15, 37, 39, 53, 159 planning method, 10, 98 structural analogy, 55–56 Hobbits and Orcs tasks, 79 Holiday Planning task, 116–118, 128 Homunculus, 92 Huntington’s disease, 205 Image, 2 Information-processing systems, 8–9 Inhibition, 85, 118, 128–129, 161, 204, 206 Inner scribe, 77, 83 Internal problem space, 4 IR, case of, 154–155

242

SUBJECT INDEX

Knowledge general, 16 plan-specific, 182 representation of, 17–18 specific, 16 Language comprehension, 3, 78 Language production, 3 Learning complex, 78 Lesions brain, 30, 79 focal, 27, 153–179, 205 frontal lobe, 31, 189–191, 209 Lobectomy temporal, 205 Lobectomy patients and problem solving, 160–161, 163, 176 Logic process models in, 75–76 Logic Theorist program, 8–9 Long-term memory, 4, 7, 44, 84 Look-ahead, 201, 203–204, 210–211, 213, 219, 222–223 underpinning of planning performance by, 204–210 Manikin test, 82 Means-ends analysis, 5, 9–10, 15, 37, 39, 53, 159 and problem reduction, 74–75 and TOH task, 204 Medial frontal cortex, 153 Memory, 3 dynamic, 18–19 episodic, 200 long-term, 4, 7, 44, 84 recognition, 44 short-term, 4, 7, 21, 84 working, 2–3, 7, 12, 17, 28, 71–88, 200, 211 Memory organization packets, 62 Mental representation, 2 Missionaries and Cannibals puzzles, 53, 79, 159 Modal model, 76

Motivation, 3 Move selection strategy, 80 Move tasks, 78–83 Multiple Errands Task (MET), 29, 122–124, 166, 175, 189–190, 217–219 effects of age, 126, 222 measure of performance, 126 as measure of plan execution, 127 simplified version of, 219 Multistep tasks, 91 Multitasking test, 219 NART-IQ, 120 Neural substrates of performance localization of, 142–145 Neuroimaging, 25, 27, 111, 130, 146, 206 during executive tasks, 153–154 functional, 142–145, 194, 207–208 Neuropsychology, 181 and planning, 26–27 Nine-dot problem, 25, 53–55, 57–60 anticipated closeness to solution of, 57 line outside/line within versions, 59–60 NOAH model, 11 Obsessive-compulsive disorder, 205 Orbitofrontal cortex, 153 Paint the ladder–paint the ceiling plan, 11 Paper and Pencil task, 127, 171 effects of age, 126 measure of performance, 126 as measure of plan formulation, 127 Parafoveal vision, 140 Parietal cortex, 208 Parkinson’s disease, 135, 143, 185, 191, 205 Party Planning task, 116–118, 127–129, 166, 220 Pattern tapping, 80

SUBJECT INDEX

Perseveration, 170 Phonological loop, 21, 76–77, 79–80, 84 Phonological store, 77 Plan plane, 12 abstraction, 12 executive, 12 knowledge-base, 12 meta-plan, 12, 73 Planning of abstract tasks, 124–125 action, 119–122 and adult ageing, 29, 45, 111–133 of appropriate solutions, 210 and the brain, 181–198 and brain damage, 30 cognitive model of, 12, 135–151 and cognitive neuroscience, 181–198 cognitive perspectives on, 182–184 computational models in, 75–76 computational perspectives on, 182–184 concurrent, 36, 38, 68 and construct validity, 220–221 and control of thought and action, 89–110 and creative expertise, 63–67 cross-cultural factors, 3 deficits, 219–220 development of strategies, 156–158 developmental perspective, 3, 107 distractibility in, 184 dynamic world, 184 effectiveness of initial, 46–48 effects of ageing on, 29, 45, 111–133 emotional factors, 3 as expectation rather than intension, 208 experimental evidence for, 211–213 and expert skill, 60–67 as expression of stored preferences, 202–204, 211 and external memory aids, 71 failures of, 191–192 global, 54, 68–69 goal-directed, 37–38 hierarchical/total order, 38, 45

243

higher order, 96–105 and ill-defined problems, 53–70 impairment, 31 initial, 36, 38, 68 investigation through simulation, 166–170 levels of, 24 literature on, 125–131 local, 54, 68–69 localization, 219–220 lower-order, 93–96 and methodology, 24–26 models, 10–12 neurological evidence for, 217–219 and neuropsychology, 26–27 non-hierarchical/partial order, 38, 45 on-line, 54, 57–60, 68 opportunistic, 37–38, 68, 182–184 optimizing performance in old age, 129–130 over optimism in, 114 partial order, 38, 68 in patients with focal brain damage, 153–179 pre-compiled, 54 problem reduction approach, 73–75 in puzzle solving, 55–60 as a range of disparate activities, 200–201 reactive, 184 for real versus abstract tasks, 125–127, 213–217 real-world, 122–124, 166, 209–211 of realistic tasks, 124–125 reason for, 105–107 state-action approach, 73–74 strategic use of, 106–107 strategies, 2, 73–75, see also Planning strategy successive refinement models, 182–183 tactics, 2 theoretical background to, 2–24 and theories of cognition, 17–20 theories of, 2 timing of, 105–107

244

SUBJECT INDEX

Planning (Contd.) tips, 2 total order, 38, 68, 184 types of, 1–2 in well-defined domains, 35–51 and willed (executive) control, 22–23 and working memory, 20–22, 71–88, 115, 128–129 as working memory for the future, 146–148 worthwhile, 105 Planning behaviour characterization of, 36–38 selection and effectiveness of, 39 Planning method, 10 Planning processes search for, 199–227 Planning research bottom-up strategies in, 10–12 top-down strategies in, 10–12, 37 Planning strategy, 2, 73–75 and individual and group differences, 44–46 and problem complexity, 39–40 and problem-solving environment, 41–44 Plans cognitive view of, 186–187 and conditional subplans, 216 and creative expertise, 63–67 defined, 183 execution stage of, 72 and expert skill, 60–67 formulation of, 205, 209 inability to follow, 205, 209 and knowledge structures, 182 laboratory formulation of, 115–119 production stage, 72 recall as expertise, 61–63 and schematized knowledge, 16–17 and structure of behaviour, 12–15 as templates, 62 types of, 1–2 understanding of, 17 Positron emission tomography (PET), 142–144, 188–189

H2O, 221 O, 188 Prefrontal cortex, 27, 45, 192–194, 208 dorsolateral (DLPFC), 153, 206, 208, 211, 220 lesions of, 173, 189–191, 205, 210 in route finding, 188 Premotor cortex, 26 Problem complexity and planning strategy, 39–40 Problem solving, 3–10 computational models in, 75–76 display-based theories of, 41 information provision in, 36 planning “what”, 96–99 planning “when”, 99–105 strategies, 45 verbalization during, 42–43 in well-defined domains, 35–51 Problem-solving environment and planning strategy, 41–44 Problem space, 2, 5 internal, 4 Problems basic structure of, 4 components of, 72 definition of, 72 ill-defined, 53–70, 72 insight, 57 non-insight, 57 semantically impoverished (knowledge-lean), 72, 78 semantically rich (knowledge rich), 72 solution plans from prior, 55–56 well-defined, 53, 72 Processes event- and time-driven, 12 Production rule, 91 Production system architectures, 7, 22 models, 7 Program design, 61–62 Programming plans approach, 61–62 Progressive deepening approach, 74 Prolog language, 62, 65

SUBJECT INDEX

Prospective memory activity-related tests, 173–174 event-related tests, 173–174 time-related tests, 173 virtual reality exploration of, 170–177 Prospective memory tasks, 127 Psychological Refractory Period (PRP) procedure, 96 Putamen, 143, 145 Puzzle solving planning in, 55–60 Puzzles goal state of, 4 ill-defined, 28 initial state of, 4 operators of, 4 restrictions of, 4 Radiation problem (Duncker’s), 25, 53–56 Raven’s Advanced Progressive Matrices, 81, 220 Reading comprehension, 78 Reading span measure, 78 Real-life tests, 171 Reasoning, 78 Recognition memory, 44 Recollection autobiographical, 216 Regional cerebral blood flow (rCBF), 221 Rehearsal process, 77 Rehearsal speed, 82 Response–Stimulus Interval (RSI), 95 Retuning, 91 Rule breaks virtual reality exploration of, 170–177 Schemas, 62, 90, 169, 182 condition–action, 22, 91 task-set, 91 Schizophrenia, 205 Scripts, 16–17, 62, 182–183, 190 Sculpting, 91

245

Selection equivocation, 162 Self-ordered Nonverbal task, 158 Self-ordered Pointing test, 119, 165 Shopping Plan Test (SPT), 213–217, 220 Shopping tasks, 25 grocery, 24 Short-term memory, 4, 7, 20, 84 Silly sentence span, 81 Single photon emission tomography (SPECT), 142, 189 Six Elements Task (SET), 29, 119–122, 128, 166, 175, 219 effects of age, 126, 222 measure of performance, 126 as measure of plan execution, 127 modified, 166, 219–220 planning aids in, 122 prospective memory paradigm in, 121 SOAR (State, Operator and Result) architecture, 10, 18–19, 43, 183 dynamic memory, 18 memory capacity of, 22 production system model, 76, 139–140 State action and search selection equivocation, 161–164 State-space diagram, 4–5 Stockings of Cambridge, 136 Stored preferences, 201 Story grammars, 182 Strategy application disorder, 189 Strategy formation virtual reality exploration of, 170–177 Striatum, 189 Stroop costs, 95 Stroop effect, 93–96, 119 Structural analogy, 55–56 Structured Event Complexes (SEC), 27 Supervisory Attentional System (SAS), 22, 26–27, 90, 92 contention scheduling mechanism of, 116, 170

246

SUBJECT INDEX

Supervisory System Model Mark II, 27 Switch costs, 94–95 Synchronization task, 100–105 Task, 91 Task environment, 4 Task switching, 93–94, 96, 105 functions, 85 Task-set, 91 Temporal lobe damage to, 138 Test of Everyday Attention, 129 Test Operate Test Exit (TOTE) feedback loops, 12–15, 91, 183 Themes, 182 Thought planning and executive control of, 89–110 Total planning framework, 184 Tower of Hanoi (TOH) test, 7, 24–25, 31, 53, 79, 154, 188 anticipated closeness to solution of, 57 five-disc version, 19–20 five-move version, 161 four-disc version, 39 goal–subgoal conflict in, 114, 159–161, 206–207, 211, 223 introduction of, 204 preparation phase of, 47 problem solving on, 158–165 role of inner scribe in, 77 size constraints of, 40 strategy changes and proficiency in, 44–45 three-disc version, 8, 37, 39, 161–162 Tower of London (TOL) task, 4–5, 7, 25, 30, 78–83, 128, 188, 204 age effect on solutions, 80–81, 112–115 cerebral blood flow (CBF) during, 142–143, 146 and cognitive planning, 135–151 complex version of, 40

computerized version of (CANTAB), 136–139 and cortex activation, 147–148 effects of age, 126, 130 eye-movement study of, 139–142 five-disc version, 28–29, 45–46, 79, 96–99, 112–113, 188 five-move version, 136 four-move version, 136 ineffectiveness of planning for, 73, 105–106 measure of performance, 126 as measure of plan execution, 127 neural network implementation of, 207 and planning processes, 200 planning processes in, 80 as prototypical task, 222 role of inner scribe in, 77 state-space diagram for, 6 three-move version, 136 time errors in, 106–107 two-move version, 136 visual and spatial requirements of, 79, 128 Tower of Toronto (TOT) task, 188 Trace decay, 21 Updating functions, 85 Ventromedial cortex, 189, 191 Verbal short-term storage, 81 Verbal speed, 82 Verbal working memory factor, 82 Verbal working memory span, 81 Verbalization, 42–43 Virtual planning task, 116–118, 129 complexity component, 167–168 context component, 167 as measure of plan formulation, 127 time specificity component, 167–169 Virtual Planning Test (VPT), 30, 166–169 Virtual Reality tests, 170 Visual cache, 77

The Cognitive Psychology of Planning (Current Issues in Thinking & Reasoning)

Working Memory and Thinking: Current Issues In Thinking And Reasoning

Rationality and Reasoning (Essays in Cognitive Psychology)

Reasoning and Thinking (Cognitive Psychology, a Modular Course)

The Psychology of Proof: Deductive Reasoning in Human Thinking

Advances in Psychology Volume 40 Current Issues in Theoretical Psychology

Psychology of proof and deductive reasoning in human thinking

Hypothetical Thinking: Dual Processes in Reasoning and Judgement (Essays in Cognitive Psychology)

Current Issues in Cosmology

Current issues in causation

Current Issues in Cosmology

Development and Reasoning: A Special Issue of Thinking & Reasoning (Thinking & Reasoning, Volume 10, Issue 2, May 2004)

Current Issues in Applied Memory Research (Current Issues in Memory)

Advances in Psychology Volume 102 Cognitive Issues in Motor Expertise

Advances in Psychology Volume 102 Cognitive Issues in Motor Expertise

Neural-Symbolic Cognitive Reasoning

Neural-Symbolic Cognitive Reasoning

Current Issues in Bilingualism: Cognitive and Socio-linguistic Perspectives

Current Issues in Bilingualism: Cognitive and Socio-linguistic Perspectives

Textbook of Cognitive Psychology

Neural-symbolic cognitive reasoning

Mental Models & the Mind, Volume 138: Current developments in Cognitive Psychology, Neuroscience and Philosophy of Mind (Advances in Psychology)

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

The Cognitive Neuropsychology of Schizophrenia (Essays in Cognitive Psychology)

The Cambridge Handbook of Thinking and Reasoning

Issues in the Psychology of Women

Issues in the Psychology of Women

The Cognitive Psychology of Planning (Current Issues in Thinking & Reasoning)

Working Memory and Thinking: Current Issues In Thinking And Reasoning

Rationality and Reasoning (Essays in Cognitive Psychology)

Reasoning and Thinking (Cognitive Psychology, a Modular Course)

The Psychology of Proof: Deductive Reasoning in Human Thinking

Advances in Psychology Volume 40 Current Issues in Theoretical Psychology

Psychology of proof and deductive reasoning in human thinking

Hypothetical Thinking: Dual Processes in Reasoning and Judgement (Essays in Cognitive Psychology)

Current Issues in Cosmology

Current issues in causation

Current Issues in Cosmology

Development and Reasoning: A Special Issue of Thinking & Reasoning (Thinking & Reasoning, Volume 10, Issue 2, May 2004)

Current Issues in Applied Memory Research (Current Issues in Memory)

Advances in Psychology Volume 102 Cognitive Issues in Motor Expertise

Advances in Psychology Volume 102 Cognitive Issues in Motor Expertise

Neural-Symbolic Cognitive Reasoning

Neural-Symbolic Cognitive Reasoning

Current Issues in Bilingualism: Cognitive and Socio-linguistic Perspectives

Current Issues in Bilingualism: Cognitive and Socio-linguistic Perspectives

Textbook of Cognitive Psychology

Neural-symbolic cognitive reasoning

Mental Models & the Mind, Volume 138: Current developments in Cognitive Psychology, Neuroscience and Philosophy of Mind (Advances in Psychology)

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

Cognitive Psychology

The Cognitive Neuropsychology of Schizophrenia (Essays in Cognitive Psychology)

The Cambridge Handbook of Thinking and Reasoning

Issues in the Psychology of Women

Issues in the Psychology of Women

Recommend Documents