Controlling Uncertainty

Decision Making and Learning in Complex Worlds

Magda Osman

• Publisher: Wiley-Blackwell
• Year: 2011
• ISBN:
  9781444351804
• Edition: 1
• Topic: Psychology
• Subtopic: Industrial & Organizational Psychology
• Index: Psychology

Controlling Uncertainty: Decision Making and Learning in Complex Worlds reviews and discusses the most current research relating to the ways we can control the uncertain world around us.

• Features reviews and discussions of the most current research in a number of fields relevant to controlling uncertainty, such as psychology, neuroscience, computer science and engineering
• Presents a new framework that is designed to integrate a variety of disparate fields of research
• Represents the first book of its kind to provide a general overview of work related to understanding control

Chapter 1

Introduction

There are two key characters in the story of control, namely, the master puppeteer and the puppet. Assume now that the master puppeteer is us, and our goal is to control the behaviour of the puppet. The puppet can represent any system that requires our control. For instance, the puppet could represent a system that is biological, like keeping our physical fitness levels up so that we can run a marathon. It could be economic, like a stock market in which we are maximizing our profit by buying and selling shares. It could be organizational, like marshalling a troop of soldiers to protect a safety zone. It could even be ecological, for instance trying to sustain an endangered ecosystem like a coral reef. More typically, when we think of control systems, what comes to mind is something industrial like operating a nuclear power plant, or mechanical like driving a car, or safety critical like flying a plane.

Clearly, then, there are many examples of control systems, some of which we are likely to experience on a regular basis in our daily lives. Control systems, therefore, are a rather broad subject and, as will be made clear in this book, can encompass almost everything. So how are we able to do it; how are we able to exert control over these various types of systems? For those who have thought of this question, this book will be a guide to some answers.

Puppets, Puppets Everywhere …

The reason that control systems can be seen everywhere is that almost anything can be thought of as a system, and most things we do in our lives are driven by our need to control. But, while the systems may be incredibly pervasive, they do adhere to some basic characteristics. Control systems are complicated. This is because they have a number of elements that will vary all at once and from one point in time to another – like puppets. Puppets can have multiple components; their moving parts (e.g., arms, legs, feet, hands fingers and head) are more often than not interconnected (e.g., fingers to hand, hand to arm etc.), and can vary all at once (e.g., performing a jump) as well as singularly in one point in time (e.g., raising a hand to wave). Thus, given these capacious characteristics of control systems, they are quite literally everywhere.

… It Takes All the Running You Can Do, to Keep in the Same Place¹

In these systems we are often required to manage the events that occur in a way that leads to something predictable, and desirable. This can be incredibly difficult to achieve and takes many years of training (e.g., becoming a pilot of a passenger jet), because the system varies of its own accord, as well as because acting upon it makes it change in some way. Or, more often than not, it is a combination of us acting on it and it doing something itself that produces changes in events. To bring the analogy of the puppet to bear more obviously with control systems, in the story the puppet had its own internal mechanism that also made it move. Imagine how hard it is to control a malleable puppet and make it dance in time to a tune without an internal mechanism that can make it move on its own. Now imagine how much harder it is when it can move on its own and not always predictably. As hard as it seems, we are capable of achieving this. So we return to the question again: how are we able to exert control over such a complicated situation?

Vladimir: ‘Say Something!’ Estragon: ‘I’m Trying. … In the Meantime Nothing Happens’. Pozzo: ‘You Find It Tedious? ’ Estragon: ‘Somewhat’.²

To answer the question ‘How? ’, we need to find a better way of asking it. First of all, finding some way of describing how these different types of systems work is of great importance, particularly if they can, on a basic level, be thought of in a similar way. Second, to complement this, our ability to control what happens in these systems should reduce to some basic psychological learning and decision-making mechanisms. They should do this because we need psychological mechanisms in place that enable us to predict the behaviour of the system and coordinate our own behaviours to effect a specific change in it. Therefore, finding some way of describing our psychological processes, along with describing the control system itself, is crucial to having an understanding of control (i.e., the scientific pursuit) and being able to improve our ability to manipulate our environment (i.e., the applied pursuit).

Given the extensiveness of both objectives, typically at the start of books like this there is a tendency to spell out at the beginning what things will not be included and what can’t be achieved. I am going to avoid this. The aim of this book is to be as inclusive as possible. If you’ve flicked through it already, you will have noticed that there are chapters spanning subject areas that include philosophy, engineering, cybernetics, human factors, social psychology, cognitive psychology and neuroscience. In order to get to the answer of ‘How? ’, we need to consider the various contributions that each of these subjects has made. The issue of control invites attention from many disciplines that don’t always speak to each other. Putting them side by side in chapters in a book is also a way of showing how they in fact do relate. Moreover, they also provide the groundwork for my answer to the question of ‘How? ’ which is presented at the end of this book.

There are two important ideas that will help to carry you along this book: (1) all the themes introduced in this book are reducible to five basic concepts: control, prediction, cause–effect associations, uncertainty and agency; and (2) all of the issues that these basic concepts raise are ultimately, and will in this book be, directed towards addressing one question, which for the purposes of this book is THE question: how do we learn about, and control online, an uncertain environment that may be changing as a consequence of our actions, or autonomously, or both?

To understand the issue of control psychologically, and to understand the control system itself in all its various guises, we have to become familiar with these five core concepts and how they are tackled through the eyes of each of the aforementioned subjects. However, I am not alone; this endeavour has been embarked on by many,³ and throughout the different chapters of the book it will become apparent that there are various ways of understanding the psychological and objective characteristics of control systems. Therefore, I will take the opportunity here to qualify why this book is not a reinvention of the wheel, by stating what it hopes to do differently.

The Aim of This Book

Role 1: catalogue

At its most humble, this book serves the purpose of being an inventory of sorts of what we currently know in a range of disciplines (e.g., engineering, artificial intelligence [AI], human factors, psychology and neuroscience) about control systems and control behaviour. Though not ever seriously taken up, an appeal of this kind was made in the late 1940s by Wiener, the self-proclaimed father of cybernetics – a discipline designed to study all matters related to self-organizing systems. Wiener (1948) hoped to bring together many disciplines to understand common problems concerning control. However, Wiener (1948) proposed that ‘the very speed of operations of modern digital machines stands in the way of our ability to perceive and think through the indicators of danger’ (p. 178). The effort in understanding all matters related to control came with a warning that technological advances may be such that artificial autonomous agents would be controlling our lives. That is, in the future the puppet would eventually rule the puppeteer, and not the other way around. Though the worry that control systems will reach a level of self-organization that may challenge our mastery of the world is perhaps unwarranted, surveying the most recent advances in theory and practice should give us a better understanding of what control systems can do, and our place with respect to them.

As suggested, an overhaul of this kind has yet to be undertaken, and so this book is an opportunity to do just that. For instance, due to the increasing complexity of the systems under our control (e.g., systems that identify tumours in X-ray images, voice recognition, predicting stock market trends, creating game play in computer games and profiling offenders), there are in turn ever increasing demands placed on them to achieve optimal performance reliably. Even something as prosaic as the car now includes an increased level of automation. This is generically classified under the title of driver assist systems (DAS). DAS now include electric power-assisted steering (EPAS), semi-automatic parking (SAP), adaptive cruise control (ACC), lane departure warning (LDW) and vehicle stability control (VSC). All of these things now influence the ride and handling of vehicles we drive. So we might ask ourselves, if we have handed over so much autonomy to the car, what control do we have?

More to the point, disciplines such as control systems engineering present us with ever growing challenges because the control systems (e.g., car) that are part of our everyday interactions continue to increase in their capabilities and complexity. If complexity is increasing, then surely we need to know how we cope with it now, especially when things go wrong. Increasing complexity in our everyday lives doesn’t just come from controlling devices such as cars. There has been a charted increase in the complexity of the decision making involved in economic, management and organizational domains (Willmott & Nelson, 2003). We can spot this complexity because some of it filters down to our consumer choices. For instance, take shopping. We have to adapt to the growing complexity that we face in terms of the information we have to process (e.g., more available product information), the choices we are presented with (e.g., more products to choose from) and the changing goals that we are influenced by (i.e., desires, aspirations and expectations). At the heart of adapting to the increasing level of complexity in our lives is our ability to still exert control. So, given the new challenges and demands that are placed on us in our lives right now, this book may be considered a sort of stock take of relevant and current research in the study of all things control related.

Role 2: solving the problem of complexity

A broader aim of the book is to help clarify what we mean when we say an environment is complex, and what it is about control systems that invites researchers from different disciplines to refer to them as complex. The complexity issue is important for the reason that there needs to be a cohesive idea about what makes control systems difficult to understand, and why we can fall into traps when we come to control them.

For instance, the term complexity has a specific reference in design engineering and is a measure of the structure, intricateness or behaviour of a system that characterizes the relationship between its various components.⁴ Thus, the properties of a control system can be specified according to objective characteristics of complexity from an engineering perspective. Efforts in defining complexity have also been attempted from a psychological perspective. Studies of human behaviour in control systems have taken properties of a control system (e.g., transparency, dynamics, number of variables, number of connections and functional forms – linear, curvilinear, stochastic and feedback loops) and investigated how competent we are at controlling systems when these properties are manipulated. The steady amassing of data from experiments along these lines started from early work by Dörner (1975). But, unfortunately, despite the wealth of findings, there has been little headway in being able to say generally what contributes to making a system complex from a psychological point of view.

This is a major problem. We simply can’t say what makes systems complex in general. What we can only say is what might make a particular system complex. This is hugely limiting because we can’t generalize to different types of control systems the psychological factors associated with them. In other words, the analogous situation would be this: we might know that the puppet in your hand is going to be hard work to operate because every joint is movable. But we wouldn’t know why it is that if, given a variety of other puppets, you still find it hard to operate them all. Is it down to psychology? Is it down to the way the different puppets function? Or is it a combination of both?

Attempts to answer these questions involve identifying possible measures of control systems complexity in order to predict the success of psychological behaviours. Again, these include borrowing ideas from another discipline, in this case computer science, which have been used to measure how controllable the system is (e.g., whether it is in non-polynomial time [NP] or Polynomial time [P]),⁵ the size of the system (e.g., its search space) and the number of interdependent processes that are contained within it. However, to date, none of these captures the true complexity of a control system, or for that matter accurately predicts what psychological behaviours we are likely to use. While it might be the case that from a computer science perspective a complicated control system can be described in some mathematical way, and from this, based on a long arduous formal process, some claims are made about how we ought to behave in order to control it, there is a problem. There is a lack of correspondence. Humans have elegant and simple means (e.g., heuristics)⁶ to reduce complexity that may be formally difficult to describe, but equally there are examples in which humans find it virtually impossible to tame a complex situation that may, from a computer science perspective, be mathematically simple to describe.

Thus, there is a gap. Knowing what defines the system as complex is no guarantee for understanding the types of behaviours needed to learn about it in order to eventually control it. Therefore, the aims of the book are to present ways of clarifying the problems faced when attempting to define complexity, and to offer a solution to it. The solution is based on describing control systems as uncertain environments. What I propose is that identifying and measuring uncertainty of the system boil down to tracking over time when changes to events in the system occur that are judged to occur independently of our actions, while also tracing those changes that are rare but have a substantial impact on what happens overall, and that also didn’t result from our actions. These descriptions of the system’s behaviour need to be integrated with people’s judgements about how confident they are that they can predict the changes that occur in a control system, and people’s judgements as to how confident they are that they can control the changes that occur in a control system. These are what I claim to be the bare essentials of understanding and developing a metric for examining the success of controlling uncertainty in control systems. However, you could argue that I’ve just performed a sleight of hand by saying complexity is uncertainty in which case I haven’t solved the problem of complexity at all. Let this might be true, but let me qualify a few things. First of all, I’m not claiming that complexity and uncertainty are the same thing. What I am claiming is that translating the issue of complexity into defining what makes a control system uncertain paves the way for connecting objective descriptions of the control system with psychological descriptions of our behaviour. Thus, uncertainty as a concept is better than complexity as a bridge between the control system and us. Second, I’m not saying complexity is reducible to the other. Instead, I am suggesting that, to answer the question of what makes a system complex a new perspective ought to be taken. From a practical standpoint, attempts to solve the mystery of what makes a system complex have thus far been illuminating but ultimately unsuccessfully in defining what makes a system difficult for us to control. I argue that it may be easier to decide on the objective properties of the control system that are uncertain than those that make it complex. This pragmatic point then, arguably, is a good starting point for presenting a framewor...

Table of contents

• Cover
• Title page
• Copyright page
• Preface: the master puppeteer
• Acknowledgements
• Chapter 1: Introduction
• Chapter 2: Causation and agency
• Chapter 3: Control systems engineering
• Chapter 4: Cybernetics, artificial intelligence and machine learning
• Chapter 5: Human factors (HCI, ergonomics and cognitive engineering)
• Chapter 6: Social psychology, organizational psychology and management
• Chapter 7: Cognitive psychology
• Chapter 8: Neuroscience
• Chapter 9: Synthesis
• Chapter 10: Epilogue: the master puppeteer
• References
• Index