NICK CHATER∗ and MIKE OAKSFORD THE RATIONAL ANALYSIS OF MIND AND BEHAVIOR

Rational analysis (Anderson 1990, 1991a) is an empirical program of attempting to explain why the cognitive system is adaptive, with respect to its goals and the structure of its environment. We argue that rational analysis has two important implications for philosophical debate concerning rationality. First, rational analysis provides a model for the relationship between formal principles of rationality (such as probability or decision theory) and everyday rationality, in the sense of successful thought and action in daily life. Second, applying the program of rational analysis to research on human reasoning leads to a radical reinterpretation of empirical results which are typically viewed as demonstrating human irrationality. Rationality appears fundamental to the understanding of minds and behavior. In clinical psychology, as well in the law, it appears to be of fundamental importance to be able to draw a boundary between sanity and madness, between rationality and irrationality. In economics, and increasingly, other areas of social science, human behavior is explained as the outcome of “rational choice”, concerning which products to buy, whom to marry, or how many children to have (Becker 1975, 1981; Elster 1986). But assumptions of rationality may go much deeper still – they seem to lie at the heart of the folk psychological style of explanation in which we describe each other’s minds and behavior (Cherniak 1986; Fodor 1987). Assumptions of rationality also appear equally essential to interpret each other’s utterances and to understand texts (Davidson 1984; Quine 1960). So rationality, in an intuitive sense, appears to be at the heart of the explanation of human behavior, whether from the perspective of social science or of everyday life. Let us call this everyday rationality: rationality concerned with people’s beliefs and actions in specific circumstances. In this informal, everyday sense, most of us, most of the time, are remarkably rational. In daily life, of course, we tend to focus on occasions when reasoning or decision making breaks down. But our failures of reasoning are only salient because they occur against the background of rational thought and behavior which is achieved with such little apparent effort that we are inclined to take it for granted. Rather than thinking of our patterns of everyday thought and action as exhibiting rationality, we Synthese 122: 93–131, 2000. © 2000 Kluwer Academic Publishers. Printed in the Netherlands. 94 NICK CHATER AND MIKE OAKSFORD think of them as just plain common sense – with the implicit assumption that common sense must be a simple thing indeed. People may not think of themselves as exhibiting high levels of rationality – instead, we think of people as “intelligent”, performing “appropriate” actions, being “reasonable” or making “sensible” decisions. But these labels refer to human abilities to make the right decisions, or to say or think the right thing in complex, real-world situations – in short, they are labels for everyday rationality. Indeed, so much do we tend to take the rationality of commonsense thought for granted, that realizing that commonsense reasoning is immensely difficult, and hence our everyday rationality is thereby immensely impressive, has been a surprising discovery, and a discovery made only in the latter part of the twentieth century. The discovery emerged from the project of attempting to formalize everyday knowledge and reasoning in artificial intelligence, where initially high hopes that commonsense knowledge could readily be formalized were replaced by increasing desperation at the impossible difficulty of the project. The nest of difficulties referred to under the “frame problem” (see, e.g., Pylyshyn 1987), and the problem that each aspect of knowledge appears inextricably entangled with the rest (e.g., Fodor 1983) so that commonsense does not seem to break down into manageable “packets” (whether schemas, scripts, or frames, Minsky 1977; Schank and Abelson 1977), and the deep problems of defeasible, or nonmonotonic reasoning, brought the project of formalizing commonsense to an effective standstill (e.g., McDermott 1987). So the discovery is now made – it is now clear that everyday, commonsense reasoning is remarkably, but mysteriously, successful in dealing with an immensely complex and changeable world and that no artificial computational system can begin to approach the level of human performance. Hence, the sentiment with which we began: Most of us, most of the time, are remarkably rational. But in addition to this informal, everyday sense of rationality, concerning people’s ability to think and act in the real world, the concept of rationality also has another root, linked not to human behavior, but to mathematical theories of good reasoning. These theories represent one of the most important achievements of modern mathematics: Logical calculi formalize aspects of deductive reasoning; axiomatic probability formalizes probabilistic reasoning; the variety of statistical principles, from sampling theory (Fisher 1922, 1925/1970) to Neyman–Pearson statistics (Neyman 1950), to Bayesian statistics (Keynes 1921; Lindley 1971), aim to formalize the process of interpreting data in terms of hypotheses; utility and decision theory attempt to characterize rational preferences and rational choice between actions under uncertainty. According to these calculi, THE RATIONAL ANALYSIS OF MIND AND BEHAVIOR 95 rationality is defined, in the first instance, in terms of conformity with specific formal principles, rather than in terms of successful behavior in the everyday world. The two sides of rationality raise the fundamental question of how they relate to each other: How are the general principles of formal rationality related to specific examples of rational thought and action described by everyday rationality? This question, in various guises, has been widely discussed – in this article, we shall outline a particular conception of the relation between these two notions, focussing on a particular style of explanation in the behavioral sciences, rational analysis (Anderson 1990). We will argue that rational analysis provides a good characterization of how the concept of rationality is used in explanations in psychology, economics and animal behaviour, and provides an account of the relationship between everyday and formal rationality, which has implications for both. Moreover, this view of rationality leads to a re-evaluation of the implications of data from psychological experiments which appear to undermine human rationality. We argue that, on the contrary, experimental evidence demands a change concerning which formal account defines the normative standard in experimental tasks. This paper thus has two linked goals. The first goal is to outline what we take to be the standard role of rationality in the explanation of mind and behavior, in disciplines as diverse as experimental psychology, animal behavior and economics – we take rational analysis to be a paradigm for such an explanation. The second goal is to draw out some of the implications of the rational analysis perspective for the interpretation of experimental data which appears to show that human behavior is non-rational. We argue, instead, that human behavior is rational, if the appropriate normative standard for that behavior is adopted. Specifically, a wide range of empirical results in the psychology of reasoning have been taken to cast doubt on human rationality, because people appear to persistently make elementary logical blunders. We show that, when the tasks people are given are viewed in terms of probability, rather than logic, people’s responses can be seen as rational. The discussion falls into three main parts. First, we discuss formal and everyday rationality, and various possible relationships between them. Second, we describe the program of rational analysis as a mode of explanation of mind and behavior, which views everyday rationality as underpinned by formal rationality. Third, we apply rational analysis to re-evaluating experimental data in the psychology of reasoning. 96 NICK CHATER AND MIKE OAKSFORD 1. RELATIONS BETWEEN FORMAL AND EVERYDAY RATIONALITY Formal rationality concerns formal principles of good reasoning – the mathematical laws of logic, probability, or decision theory. At an intuitive level, these principles seem distant from the domain of everyday rationality – how people think and act in daily life. Rarely, in daily life, do we accuse one another of violating the laws of logic or probability theory or praise each other for obeying them. Moreover, when people are given reasoning problems that explicitly require use of these formal principles, their performance appears to be remarkably poor. People appear to persistently fall for logical blunders (Evans et al. 1993), probabilistic fallacies (e.g., Tversky and Kahneman 1974) and to make inconsistent decisions (Kahneman et al. 1982; Tversky and Kahneman 1986). Indeed, the concepts of logic, probability and the like do not appear to mesh naturally with our everyday reasoning strategies: these notions took centuries of intense intellectual effort to construct, and present a tough challenge for each generation of students. We therefore face a stark contrast: the astonishing fluency and success of everyday reasoning and decision making, exhibiting remarkable levels of everyday rationality; and our faltering and confused grasp of the principles of formal rationality. What are we to conclude from this contrast? Let us briefly consider, in caricature, some of the most important possibilities, which have been influential in the literature in philosophy, psychology and the behavioral sciences. 1.1. The Primacy of Everyday Rationality This viewpoint takes everyday rationality as fundamental, and dismisses the apparent mismatch between human reasoning and the formal principles of logic and probability theory as so much the worse for these formal theories. This standpoint appears to gain credence from historical considerations – formal rational theories such as probability and logic emerged as attempts to systematize human rational intuitions, rooted in everyday contexts. But the resulting theories appear to go beyond, and even clash with, human rational intuitions – at least if empirical data which appears to reveal blunders in human reasoning is taken at face value. To the extent that such clashes occur, the advocates of the primacy of everyday rationality argue that the formal theories should be rejected as inadequate systematizations of human rational intuitions, rather than condemning the intuitions under study as incoherent. It might, of course, be granted that a certain measure of tension may be allowed between THE RATIONAL ANALYSIS OF MIND AND BEHAVIOR 97 the goal of constructing a satisfyingly concise formalization of intuitions, and the goal of capturing every last intuition successfully, rather as, in linguistic theory, complex centre embedded constructions are held to be grammatical (e.g., ‘the fish the man the dog bit ate swam’), even though most people would reject them as ill-formed gibberish. But the dissonance between formal rationality and everyday reasoning appears to be much more profound than this. As we have argued, fluent and effective reasoning in everyday situations runs alongside halting and flawed performance on the most elementary formal reasoning problems. The primacy of everyday rationality is implicit in an important challenge to decision theory by the mathematician Allais (1953). Allais outlines his famous “paradox”, which shows a sharp divergence between people’s rational intuitions and the dictates of decision theory. One version of the paradox is as follows. Consider the following pair of lotteries, each involving 100 tickets. Which would you prefer to play?