Reviewed by
Edmund Chattoe
Department of Sociology, University of Oxford
"As Wittgenstein said, when put on an unbalanced diet of examples philosophy suffers from deficiency diseases." - Geach, P. T. (1960) 'Ascriptivism', The Philosophical Review, 69(2), April, p. 221.
This twenty-chapter collection brings together psychologists interested in how past experience shapes decision-making. In particular, the book takes as it central theme the idea that feedback from the environment allows decision-makers to create "routines" in memory that facilitate and speed up decision-making in the future. The four main sections of the book cover theoretical frameworks for modelling and understanding routines, basic and applied research on routines and how training might be used to enhance routine-based reasoning in aspiring "experts".
The book is well produced and edited. Almost all of the contributions are clear, broadly relevant and worth reading. The introduction and conclusion by the editors genuinely do serve to place the rest of the book in a useful context. As such, this book is definitely an advertisement for what can be achieved by an edited volume based on a conference. In addition, the contributors clearly know the literature and the book is a very useful source of references to the wider context of psychological research on decision-making for those of us outside the field.
My curiosity about the book came from an interest in dynamic decision-making and a sense that psychology has been relatively neglected as a source of ideas in social simulation. I think this book raises some very interesting questions that deserve wider consideration in our field but, understandably, provides only partial answers. It is reasonable then to ask both what the ideas in this book might do for simulation and what simulation might do for the research programme which the book proposes. I shall discuss these points in order.
The first thing that struck me on reading the book is that the research programme is clearly still gelling. In a compilation, it is never completely clear whether the contributors genuinely share the unified vision suggested by the editors. On the other hand, differences in use and interpretation may reflect intrinsic challenges in the material under study that deserve to be aired. At times, the book explores what seems to me the most interesting and novel meaning of "routine". This is the idea that as well as learning how to achieve a certain goal, we simultaneously learn about the structure of the problem (and perhaps how to represent or abstract it) in a way that makes achievement of the same goal easier, more effective or quicker in the future. This seems to be the meaning of routine used by the editors in their framing chapters. However, some of the individual contributions use the term (whether by accident or design) in a sense that is less interesting because it is already something of a commonplace in the bounded rationality literature. This is the idea that a routine may simply be a rule or an action (whether in general or as an outcome of choice). For example, Betsch (p. 40) proposes:
"The representation of a decision problem p consists of goals or desires (e.g., "I'm hungry, I want to get something spicy to eat") and typical context features (e.g., "It is lunch time and I am at the office"). Goals and context features provide the antecedent conditions for choice. … The routine is the behaviour (e.g., "go to the Chinese restaurant), which, by virtue of past experience, is known to serve the goals embedded in the particular type of situation."
In this description it is not clear that there is (or has necessarily ever been) a choice involved or that the term "routine" adds anything in describing the outcome beyond use of the word "action" or "rule". This quote seems to hide the most interesting part of the process which is how going to the Chinese restaurant became the relevant behaviour given the goals and context. If this research programme is not going to explore that question, it seems that adequate tools (like classifier systems or neural networks) already exist to tackle translating static representations of experience into actions.
By contrast, in the introduction, the editors describe a process in which decision-makers acquire and evaluate routines for solving different problems. This seems to be an insight that would add realism to simulations involving decision-making in complex environments. Within this framework a number of interesting corollaries immediately follow. An expert is someone with a wide repertoire of effective routines in a particular domain and a learned ability to match routines with problems quickly and accurately. By contrast, in a novel situation, the decision-maker must make a considerable effort and "feel their way" not only deliberating about how to achieve the goal but also reflecting on what can be learned concerning the structure of the problem. In a group, different members may classify the same problem in different ways and/or have different routines for solving it. This can either improve or impair group decision-making when compared to individual performance. Other corollaries not stressed include the possible path dependence of decision-making: actors are likely to start by applying the (potentially unique) set of routines they have to any new problem even though there may be a much better routine they simply do not know. (This risk has a resonance with the "theory imperialism" that is sometimes found in social science and is captured by Abraham Maslow's famous dictum: "To the man who has only a hammer, everything looks like a nail.") Furthermore, at least some kinds of teaching may involve the transfer or routines by demonstration and the diagnosis of ineffective or incomplete routines by the teacher. This meaning of "routine" seems to give rise to a productive framework but the contributors sometimes struggle to operate within it and the interest of their contributions suffers as a result.
I think one of the issues giving rise to this difficulty is the shortage of really good test cases both in the world at large and certainly as presented in the book. In order to understand routine formation and use, it seems that one needs a challenging decision problem and also a research method that does not "push" the subjects into dealing with it in a single way. Many controlled experiments are simply too stripped down to give "free" routine formation and evaluation much purchase. Some of the most interesting chapters in the book are those that either study experts "in the wild" carrying out fire fighting or air traffic control tasks or develop novel methods to elicit routines. In a particularly imaginative contribution Omodei et al. discuss how head mounted video cameras can be used to help experts "talk through" their routines when reviewing the footage. This approach both avoids contaminating the behaviour (experts rapidly forget they are wearing the cameras when dealing with real problems) and solves the recall problem (that it is extremely hard to reconstruct thought processes after the event). By contrast, the theoretical chapters seem to deal with relatively simple decisions (cast in terms of game theory for example) or to have no specific problem in mind at all. Several papers also make use of "micro-world" simulation (actually gaming) in which experimental subjects are asked to do plausible tasks – like stabilise the output of an imaginary but realistic sugar factory. However, the experimenters can embed these tasks in a sufficiently complicated environment that there is more than one way to solve the problem and it become possible to observe and even prime the routines that are used by subjects. Such techniques seem like an extremely valuable adjunct to simulation where realistic models of decision-making are needed for a particular domain. Generally simulation seems to have avoided the difficulties of understanding dynamic decision-making processes and lifted solutions "off the shelf" from economics and Artificial Intelligence. This book may thus present an interesting alternative.
The second difficulty I think the book faces is in some of the available terminology and this is to some extent linked to the problem of good case studies. I have already suggested that in order to be interesting, routine must mean something more than simply rule, action or choice. In the same vein, the distinction between deliberation and intuition (used by several contributors) is potentially problematic. Using the approach sketched in the opening chapter, intuition appears to be that process by which a particular problem is "matched" to a routine that will generate an appropriate decision. An expert has fast and reliable intuition of this kind while a novice may have no intuition or at best a faulty one (perhaps inherited from a superficially similar problem). However, it is not clear from the examples given how opposed intuition in this sense (rather than its "mystical" namesake "woman's intuition") really is to deliberation. The intuitions that strike one are in a sense "crystallisations" of previous deliberations, evaluated with deliberation and even these may still require active processes of reasoning during implementation. (The best example of routines that I could devise was dismantling a complicated piece of electronic equipment like a computer in order to replace a component. The first time this is done, there is much guessing and reasoning about which screws to loosen first, how the components actually come loose and so on. After taking a particular computer apart, one has not only converted a lot of those potential choice points to a script but one has started to infer general principles – "Don't force it. If you need to do that, you are doing it wrong." and "Work out which way it was put together and reverse the order to take it apart" – that can be applied generally.) It is not the case that a decision must either be "worried through" from scratch on each occasion or become completely automated as the dichotomy between deliberation and intuition implies. The final conceptual concern I have is that copious good case studies reveal other ways in which decisions may be challenging that may turn out to be important. For example, in expert decisions, the expert must sometimes not only implement a routine but also delegate parts of it to individuals with different abilities and motivations and evaluate the success of this delegation and the ongoing execution of the routine. It may be that this delegation task has little bearing on the choice, implementation and evaluation of routines but such issues are lost when simplified examples are used. Analysis of good case studies thus serves to define the scope of the research agenda in a way that it is not too sensitive to the strengths and weaknesses of prevailing theories.
A final set of concerns is not with the book per se but with what it implies about how the field seems to be developing to an outsider. Firstly, the theoretical frameworks presented are extremely diverse and complicated and seem to be a long way from what existing stylised facts or presented experimental settings can effectively falsify. Generally speaking, the field appears to be quite "theory heavy" which seems a shame when the novel methods presented are so interesting. (Matters are made worse by the fact that several of the chapters in the "research" sections are actually literature reviews or general discussions.) Secondly, sociology talks disparagingly of "the usual suspects", attempts to explain everything in terms of class, gender and ethnicity. In the same way, it seems that the "usual suspects" of psychology may be creeping in to make the "routines" research agenda potentially unmanageable. While not denying that factors like mood, affect and motivation have a bearing on decision-making, it seems reasonable to start by understanding routine formation and implementation more fully and only then to add complicating factors. (Furthermore, it seems to this reader that the papers treating routine as a "variable" to be influenced by other variables have the least interesting points to make.) Finally, although this is a speculative point, if we take the "routines" notion seriously, to what extent does this undermine frequent attempts in the book to present general schemas for decision-making? Just as one problem may be seen "as a game" (in the game theory sense) and another as susceptible to a set of classifiers, so might the routines and sequence for collecting supporting information and evaluating outcomes vary from setting to setting or routine to routine.
It seems to me that the ideas and methods of this book should be interesting to simulation because of its current tendency to take decision-making "off the shelf" and not to collect relevant empirical data about this aspect of behaviour for particular domains. In turn, however, what can simulation do for the "routines" approach? There seem to be two main answers to this question. Firstly, social science (and thus social simulation) has a ready supply of "real decisions" that have important ramifications and are thus worthy of study. Two examples from sociology are decisions about how much education to consume and how to participate in the labour market. At the same time, simulation is an ideal technique for translating abstract schemas into detailed mechanisms with consequences that can be tested against available data. In fact, two chapters in the collection (Chapter 1 by Johnson and Busemeyer and Chapter 5 by Sedlmeier) use computational approaches. However, the first chapter simultaneously suffers from a simple decision problem and a decision-making algorithm that looks too baroque and arbitrary to be supported by it while the second chapter alludes to a simulation rather than actually describing it. In both cases, the "social simulation style" of design and presentation would have helped these contributions and there is generally little sign that any of the authors are aware of the wider simulation community and its methodological progress.
To sum up, this is an interesting and thought-provoking book well worth reading for new ideas (and particularly for discussion of research methods novel to simulation) but it is very far from a "recipe book" of well established techniques. The jury is out on whether the "routines" approach to decision-making will coalesce and grow or fragment and collapse but the ideas are still worthy of consideration even in their current form as long as the reader expects good questions for the simulation "melting pot" rather than ready answers.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2005