CHAPTER 1 A CONVENTIONALIST THEORY OF SOCIAL RULES AND OBLIGATIONS

Transcription

1 CHAPTER 1 A CONVENTIONALIST THEORY OF SOCIAL RULES AND OBLIGATIONS 1.1. Rules, Reasons, Values, Virtues All human interaction is guided and constrained by social rules. Children in school classes, patients and nurses in hospitals, directors participating in board meetings, even visitors to a party, generally know what they are supposed to do and not to do, and generally act accordingly, largely without effort. It is at least initially plausible to suggest that law is a body of social rules, whatever else it is. But what exactly is a social rule? To approach this question, let me note some typical characteristics of interaction guided by social rules, without claiming that any of them is either a necessary or a sufficient condition. Most basically, following a social rule is a form of intentional behaviour for which the agents can be held to be responsible, even if they have internalized it to the point where it becomes a matter of routine. If following a rule is a form of intentional action, it should be possible to explain it in terms of the reasons the agents have for conformity. In the second place, social rules seem to be related to the point of the interaction. They usually make it easier for participants to achieve a common or shared aim: children being taught, patients being cured or cared for, boards making good decisions, or at least a decision, visitors to the party feeling at ease, or confirming their self-image of sophisticated people. Social rules facilitate the achievement of these aims by adjusting people's actions to each other. At the very least, they enable people to use their energy and resources efficiently, and not for managing mutual embarrassment. Without those rules, things would go worse than they do now, at least from the point of view of the people following the rules. 1 In the third place, if someone violates the rule, the others will usually respond by feeling disapproval and expressing such feelings in various ways, ranging from subtle body language to molestation and imprisonment. What is conveyed by these expressions is not only that the person is doing or omitting something without good reason. Rule-conforming behaviour is something which the person "owes" the others, which they have a "right" to expect, though these expressions need not always have strong moral connotations. They think less of him for disrespecting their rights; if his behaviour cannot be justified or excused, it reflects badly on his person. He will be condemned, not so much as irrational but rather as rude or ruthless, sly or perfidious. To summarize: interaction guided by social rules is a form of intentional behaviour, by which people coordinate their actions in order to enable themselves to 1 I do not deny that following rules often has suboptimal results, see ch

2 2 CHAPTER 1 achieve shared aims; conformity and non-conformity are interpreted in terms of personal character. If we are to understand what it means to follow a social rule, we should be able to account for those features. "Duties" should be understood by reference to "reasons", "values" and "virtues". How are these elements related to each other? In particular, how can we understand the agents' reasons in terms of the relevant values or virtues or both? That is the question I want to consider in this introductory chapter Hart on Social Rules The concept of a rule or norm I will use the concepts interchangeably can be used on either of two levels. Firstly, on the propositional level. The rule describes a possible state of affairs, but it has a peculiar "direction of fit": if its conditions of application are realized but the world fails to match the rule, this is not to be considered a defect in the rule, but rather in the world. 2 Things having this direction of fit (let us call them "blueprints") can be divided into two classes. Having compared the world to a blueprint of the first class ("values", "the good"), one judges to what extent the blueprint is realized, on a continuous scale from zero to complete realization. But in the case of blueprints of the second class ("prescriptions", "the right"), the decision one has to make is a binary one: either the prescription is realized or it is not. Of course a prescription can be vague, in which case it is, within a certain range, difficult or even impossible to decide whether it is realized or not. But the point is that the decision which it is difficult or impossible to make within this grey zone is the binary one. Prescriptions tend, directly or indirectly, to refer to actions, more so than values which refer to other states of affairs as well. A prescription may be particular, picking out a token action. Or it may be universal, describing certain conditions in general terms, and stating that whenever these are satisfied, a specific description should be satisfied as well. A rule is a universal prescription, governing acts. 3 A rule, however, is not only an element in Popper's third world, but also in his second: it is a psychological and/or social reality. For it is used by people to guide their behaviour. It may be used in this way by one person, or by a number of (in this respect) unrelated persons. In that case, it is a private rule. Such rules may be considered as standard plans, which have the function of saving deliberation costs in recurrent situations requiring action. 4 A social rule, on the other hand, is used by a number of people for guiding their behaviour together. It is this "together" which causes problems of explanation. One of the more ephemeral aspects of Wittgenstein's influence on philosophy was the astonishing popularity of the notion of a rule during the sixties. It was 2 Anscombe 1957, 4, 56; Kenny 1963, 216-7; Searle 1983, "A function which can take an indefinite variety of decision-types as inputs and deliver in each case one option or set of options as output..." Pettit 1990, 3. I do not want to suggest that we can identify the function without reference to the psychological and/or social reality of the rule being followed. 4 Bratman 1987.

3 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 3 considered a decisive step in the direction of a better understanding of all kinds of human practices playing games, promising, speaking a language generally, psychoanalysis, law, and even human action as such to view the subject under investigation as rule-governed behaviour. 5 The crucial features of such conduct were usually considered to be the following. Firstly, the meaning of present action is explained in terms of options and directives for future conduct, just as the meaning of a chess move can be identified with the way it determines the possibilities for the rest of the game. Secondly, rules are reference points for a group of people interacting with one another. Both features were thought to be connected: the right continuation is the one which the group considers right. 6 Now it may be true that in deciding which move is required by the rule, we have no final criterion but the shared judgment of a group of people. Following a private rule thus turns into a public practice, but not yet into a social one. This rule addiction culminated in the discovery of so-called institutional facts. 7 That Mr. Kok is the prime minister of the Netherlands, that AC Milano won the Champion's Trophy, or that I was fined yesterday for a parking offence, are surely facts, but they are only facts in virtue of the existence of a social rule. An intelligent being from another planet would not be able to observe these facts. If one tries to describe such a fact without reference to the rule, the description will only present the (relatively) brute fact; if the rule is then applied, this fact is found to have an institutional meaning. In this way, it seemed, we might even account for the binding force of promises without having to face the problems of either the volitional or the reliance theory. 8 If you solemnly utter the words: "I herewith promise to pay you, Jones, the sum of fifty guineas" (or, to give an even more brute description, if you produce the corresponding sounds), you undertake an obligation to pay Jones the sum of fifty guineas, for a (social or private?) rule exists providing that, whenever you utter these words, you undertake the obligation. 9 The problem is, of course, that the theory fails to explain what kind of fact the existence of a rule is supposed to be. As soon as we try to understand this, we may have to bring in either the concept of reliance or the concept of volitional commitment, which leads us back to the problems of the original theories. 5 E.g. on speech acts: Searle 1969; Tugendhat 1976, lectures 14 and 15; Wunderlich 1978; on psychoanalysis: Mooij 1982; on law: Hart 1961; on action: Melden 1956; Winch 1958, 52. For the limitations of this approach, see Coyne Winch 1958, 24-33, interpreting Wittgenstein's "private language" argument. 7 Anscombe 1958; Rawls 1955; Searle 1969, According to the volitional theory, the will somehow binds itself, cf. Raz 1972, Robins This is a rather mysterious act. See ch. 2.8 at note 61. According to the reliance theory, a promise binds in virtue of the fact that it would be wrong to let down the person relying on it, cf. MacCormick 1972, Atiyah 1981, MacMahon But it seems she rather relies on it in virtue of the fact that it binds. See Den Hartogh Some philosophers ask, How can making a promise create an obligation? A similar question would be, How can scoring a touchdown create six points? As they stand both questions can only be answered by citing a rule..., Searle 1969, 35, cf. Searle 1964; Hart 1961, 42.

4 4 CHAPTER 1 Similar comments can be made about other institutional accounts, for instance about the so-called institutional theory of law. 10 I do not want to deny that this theory is true; indeed, I am in complete agreement with its basic thesis that legal facts are facts in virtue of the existence of legal rules. But I find the thesis completely uninformative as long as I have not been told what kind of thing a legal rule is. In the meantime, the theory strikes me as a marvellous example of explaining obscurum per obscurius. It is true, however, that almost from the beginning attempts have been made, within this tradition, to explain what it means for a rule to exist. Peter Winch made a start in The Idea of a Social Science (1958), and the most thorough analysis has been presented by Herbert Hart in The Concept of Law (1961). The basic notion in Hart's theory of law is that of a secondary rule. In his view, law has developed from a more primitive system of social rules, by the introduction of "rules about rules", rules about identifying rules of the system, about making and changing them, and about establishing authoritatively what they require. His first task, therefore, is to tell us more about this more primitive system. That is how he comes up against the question what it means for a social rule to exist. This analysis is of central importance to my concerns. For the points about legal philosophy which I wish to make in this book can be seen as so many amendments to Hart's legal theory. And these amendments follow from the need to remedy the defects of Hart's analysis of social rules. The following inventory of conditions which determine whether a rule counts within a particular community can be elicited from the writings of Winch and Hart. 11 (1) Some order or regularity in the behaviour of a number of people (in any case more than one) must be ascertainable. I begin with this condition in order to develop the contrast between rule and regularity in the subsequent ones. (2a) Deviant or irregular behaviour occurs. (2b) Regular conduct can be learned. These two features characterize habitual behaviour. They are to some extent related; if so desired, habitual behaviour can be changed, given the necessary exercise and effort. (Hereditary behaviour can also sometimes be changed: breathing cannot, but thumb-sucking certainly can). However, these conditions do not as yet suffice: for example, someone constantly using the same catch-phrase is not usually following a rule. The next characteristics should permit us to distinguish between rules and habits, the latter being merely a product of rote learning. (3a) Those involved are capable of discerning deviant behaviour MacCormick & Weinberger 1986, Ruiter 1993, Morton Recently, some of these authors have moved in the direction of a conventionalist analysis of social norms, cf. e.g. MacCormick Interestingly, MacCormick has also proposed one of the most sophisticated versions of the reliance theory of promising: MacCormick Winch 1958, 28-33, 57-65; Hart 1961, 9-11, 54-56, 86-88, 96, Cf. also the inventories by Duintjer 1977, 26-40, and Gumb 1972,

5 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 5 (3b) In general, a negative attitude is adopted toward observed deviant behaviour; this attitude is expressed either verbally or non-verbally. (Positive attitudes to conforming behaviour are possible but not necessary.) There is, therefore, not only an overall convergence of behaviour, but also of judgment. People who thus harmonize in conduct and judgment, together make up the core of the community in which the rule holds. (3c) Those who deviate, tend in general to react to negative responses by revising their behaviour (or to justify or excuse it, etc.). The required negative response need not come from others; social correction and spontaneous self-correction coexist side by side. Hence it is not necessarily the negative attitude of someone else which gives the sinner reason for changing course, it may (also) be deviance itself. (3d) Just as awareness of transgression provides the transgressor with reason for selfcriticism, so too it provides others with a reason for criticism. Their "negative attitude" is, therefore, not only a usual and hence predictable reaction, it is also considered justified in their own eyes. (3c and 3d are Hart's particular contribution.) In short, a habit is a convergence in behaviour caused by previously converging behaviour. It is like a tidal wave: water runs into the nooks and crannies caused by a former flood. By contrast, a rule is a pattern of conforming ways, which repeatedly sanctions those involved to pursue their conformity. It is a deliberate unanimity of conduct and judgment. We are not conditioned, we intentionally allow ourselves to be led. If we compare this account with my phenomenological exercise in section 1, we recognize two elements: "reason" (3d) and "virtue" (3b). The element "value" is lacking, and the relations between the elements are not extensively explored. (Though (3c) makes a very important point which we will return to.) In particular, Hart carefully refrains from saying anything about the possible reasons agents may have for joining the queue. He only remarks that the reasons might be of any kind: "fear, inertia, admiration of tradition, long-sighted calculation of selfish interests as well as recognition of moral obligation". 13 He has his reasons for not wanting to be more specific. But the result is an important lacuna in his account which has been pointed out by Warnock. 14 Situations can develop which meet every proviso of the analysis proposed by Winch and Hart, while nevertheless there is absolutely no question of the existence of a social rule. Take the following case. 12 For Winch 1958, 57ff this is the essential difference between rules and habits. 13 Hart 1958, 93; see ch. 8.2, note Warnock 1971, (with an equally parochial example: going out without an umbrella in British climate conditions), cf. Dworkin 1986, 145; Postema 1987d, 99; Marmor 1998, ; Shapiro 1998, Hart actually recognizes the interdependent character of the relevant reasons at several places, cf. Hart 1961, 55, (the rule of recognition has to be regarded by the courts "as a public, common standard of judicial decision, and not as something which each judge merely obeys for his part only"). But in Hart 1961 he did not systematically take it into account, as he acknowledges in his reply to Dworkin, Hart 1994, , cf. Green 1996.

6 6 CHAPTER 1 Usually, groceries which easily spoil, are kept cool. Someone who keeps old newspapers inside his fridge, while leaving meat and milk outside it, deviates from this "norm", thereby leaving himself open to negative criticism. But if we ask for further specification of the criticism, we will reject an answer stating that it happens to be the rule that meat and milk are placed in the fridge; we will reject it not merely as insufficient, but rather as irrelevant. It is simply stupid to allow food and drink to go bad unnecessarily. What is lacking in such counter-examples is that the actions of people following a rule have "a reference to each other". 15 It is insufficient simply to expect the same performances from everybody and to consider deviance a ground for criticism. After all, the reasons for expected conduct might count individually for everyone, irrespective of the doings of others. Talk of a rule is then out of the question. What is required for a social rule to be a social rule, is that each member of the group has reason to maintain a certain regular pattern in his conduct if and only if others (or enough of the others) do the same, and for the same reason. That's why, if asked for a reason to do as the rule requires, we are inclined to answer "because it is the rule", unenlightening as this reply may be. Everyone is required to take his or her place in a desired pattern of interconnected actions Interdependent Reasons and Game Theory So it appears that we should concentrate our attention on this idea of the interdependence of reasons. A rule exists when people intend to act in a certain way, only because they believe the others to intend to act in the same way as well. Our next question is: in what sorts of situations are such interdependent intentions intelligible? Here our phenomenological thumbnail sketch points the way: we must look for situations in which people's actions have to be coordinated in view of some desirable outcome. But if that is the case, the situations we are looking for are "games" in the sense of the theory of games: situations in which the outcome for each of the "players" depends on the choices made, not only by himself, but also by the others. At this point David Lewis made intellectual history by showing how, in one particular situation of interdependent outcomes the so-called Coordination game the participants could have reason to make their choices in such a way that a pattern of interdependent choices emerges: a convention. My leading question will be whether this type of analysis can be extended to other "games", for instance the so-called Prisoner's Dilemma. As we will see, the problem here is this: in a Coordination game, people have 15 Hume, Treatise, III.ii MacCormick 1981, 41, proposes to complement Hart's conception of social rules in this sense. Hart 1982, ch. 10, rather follows Raz in analysing law not in terms of social rules, but of exclusionary (peremptory) reasons to follow authoritative instructions. This conception even more clearly fails to account for the interdependence of reasons. It is also inconsistent with his own account of the defeasibility of legal norms, as I argue in ch. 11, cf. Chapman 1998b.

7 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 7 reason to make their own choice fit into the pattern of interdependent decisions; in Prisoner's Dilemma games and others, however, they seem to have reason rather to deviate from the pattern which makes it difficult to understand how the pattern can either emerge at all, or be sustained once it has emerged. At that point we will have to bring in the other element mentioned in section 1, the element of virtue. But let me first explain what I think I am committed to by the use of this analytical apparatus, for this often gives rise to misunderstanding. In the first place, I am committed to assume intentional agency. A person's actions can normally be understood by the way in which they fit into the structure of his desires, beliefs, and intentions. 17 Needless to say, this is only a default assumption: I do not want to deny the occurrence of utterly irrational behaviour. In the second place, I am committed to ontological (rather than methodological) individualism: beliefs, desires and intentions belong to individual minds; there is no such thing as a collective mind. Social reality is to be understood by the way in which the beliefs, desires and intentions of individuals depend upon each other. 18 Hence we have not really explained people's behaviour by showing that it is governed by a rule; we should go on to explain why the rule is followed. In the third place, I am committed to an assumption about people's desires: that at least many of them have a maximizing structure. If a person has wants or desires, she can order states of affairs according to the extent to which they fulfil her wants; she has a use for the relational concepts 'better than' and 'as good as'. Decision theorists tend to say that she has a preference ordering; the concept of an 'ordering' implying at least that these relations are reflexive, transitive and continuous. I insist, however, that the concept of a preference ordering tells us almost nothing about the nature or content of these desires; it merely specifies some elementary structural characteristics, in particular the fact that, within the general class of "blueprints", these contents have the nature of values rather than prescriptions. That is why I said they had a maximizing structure. 19 Apart from this, nothing is specified: the desires may be egoistic, altruistic, or impersonal, they may be outcome-oriented or action-oriented, agent-neutral or agent-relative, forward-looking or backward-looking, and so on. Nor does decision theory, in introducing the concept of a preference ordering, say anything about the relation between desire and goodness: whether things are good because they are desired, or desired because they are good. It merely assumes that quidquid appetitur, appetitur sub specie boni (whatever is desired, is desired as good), it explains the maximizing structure of the good, and articulates formal conditions for the coherence of judgments in which those basic relational concepts are used: the axioms of decision theory Davidson 1980, essay 12; Dennett 1978, essays 1, 12, 14; Elster 1979, I assume that intentions cannot be reduced to beliefs and desires, cf. Bratman 1987, ch This type of individualism should not be confused with "atomism", if this is taken to mean what Nozick (1977), 1997 refers to as Crusoe-theory: the explanation only refers to beliefs, desires and intentions a person could have if he were the only one to have such mental states. This absurd theory is, of course, inconsistent with the idea of the interdependence of reasons, and with strategic concepts generally. See Pettit 1993, chs. 3-4, for a persuasive argument in favour of individualism and holism, against collectivism and atomism. 19 Broome 1991, ch. 1; Gauthier 1986, ch See any introduction, e.g. Kreps 1990; for dynamic decision theory McClennen 1990, in particular ch. 7. Some of these axioms are controversial, e.g. the separability or modularity axiom, see ch. 2.8.

8 8 CHAPTER 1 If desires have a maximizing structure (we want to realize more of the good rather than less), and if the basic ordering relations satisfy some conditions of coherence, it is possible to construct a utility scale for each agent. Again, by introducing the concept of 'utility', whatever its historical connotations, nothing is specified about the contents of desire or the nature of the good. Goodness, and the satisfaction of the correlated type of desire, can be measured on a continuous scale; this scale measures "utilities". 21 Though it may be hard to believe, decision and game theory are equally compatible with neo-aristotelian virtue ethics as with micro-economics. Note that I do not claim that all desires have a maximizing structure. On the contrary, the very concept of a rule requires that at least the desire to follow a rule has a binary structure: it is either satisfied or it is not. Some of our desires make, or recognize, requirements on us. I also recognize that even for desires with a maximizing structure, it may not always be possible to measure their aggregated satisfaction on one continuous scale: they may be incommensurable. The only claim I make in this connection is that in many situations in which people are engaged in common enterprises governed by rules, the subset of their relevant preferences satisfies the condition of commensurability. My use of the concept of 'utility' seems to presuppose the Humean conception of reasons being derived from beliefs and desires. But this conception misrepresents the internal point of view of the deliberating subject. It is not her belief which provides her with a reason, for if the belief is false, she may not actually have the reason she believes she has. Hence the true reason is not her belief, but the fact which she thinks she has apprehended. 22 These facts need some normative background in order to be relevant to Practical Reason. Here a similar move from the subjective to the objective seems to be required by the agent's internal point of view: it is not her desire as such which provides the necessary background for some fact to be a reason, for the desire may be criticizable. However, the court of appeal for actual desires can either be thought of as consisting of the agent's pattern of desires as a whole, or of a system of objective values, or of a mixture of both. (Alternatively, the objectivity of value can be thought of as being constituted by an ideal harmony of desire.) As I indicated already, I do not want to commit myself to any particular view on the relation between value and desire. But two points have to be recognized: on the one hand that specific desires can fail to invest some facts with the force of reasons because they are criticizable, on the other that the same fact may provide a reason for one agent but not for another because of their different motivational make-up. 23 For the external point of view, the Humean model is adequate. 24 We explain people's intentional actions by reference to their actual beliefs and desires, even if the 21 This is an interval-scale, with an arbitrary zero-point and unity, allowing for positive linear transformation of the numbers, as e.g. from Fahrenheit to Celsius. It cannot be used for making interpersonal comparisons. 22 See ch To that extent, the Humean model should not be completely rejected. 24 Some authors have argued that beliefs may motivate without the presence of desire, e.g. Nagel 1970, Platts 1980 and, most radically, Dancy 1993.

9 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 9 beliefs are false and the desires open to criticism. 25 Many if not most economists, if pressed, will still assert their formal adherence to the so-called revealed preference conception of utility, according to which we only ascribe beliefs and desires to people on account of the actual choices they make, in order to predict other actual choices they will make. Why people should be expected to behave as if they were informed by reasons, if as a matter of fact they need not have the corresponding beliefs and desires, is one of the deeper mysteries of economic science. The economists may protest that they are only interested in coherent choice, but the point is that their understanding of coherence makes no sense outside the framework of the belief/desire model of folk psychology. However that may be, if it is true that social rules must be understood in terms of interdependent reasons, we will have no use for the behaviourism of revealed preference theory in that enterprise. It will be clear that I am prepared to permit myself a less restricted diet of motivational assumptions than is usually associated with rational-choice theory. I am certainly not postulating homo economicus. This may raise the suspicion that the gametheoretical framework has no useful work to do in my theory. If I am ready to ascribe additional motives to people whenever I run out of the usual motivational assumptions of rational-choice theory, my explanations may seem to smell strongly of virtus dormitiva: people follow norms because they have a disposition to do so. I hope to convince the reader that my explanations are more interesting than that. Indeed, it is one of my aims in this book to show that the theory I develop really makes a difference in this case a difference to the solutions proposed for some perennial problems of legal philosophy. And if it makes a difference, it cannot be vacuous. Although I am quite tolerant about the nature and content of the preferences which set up the interaction patterns ("games") I want to discuss, one might wonder whether my concept of rationality does not still represent the reductionism usually associated with rational-choice theory. And, indeed, the concept of rationality I will use is a purely instrumental one: the rational choice is the one which maximizes utility. But again, I do not claim that this is all there is to being rational. I have already made room for the rational criticism of desires. The only claim I make is that in so far as you have rationally acceptable preferences with a maximizing structure, it is unquestionably rational, other things being equal, to decide for the option which in the long run maximizes your utility (leaving aside the problem of time-discounts). 26 That too seems a quite harmless, indeed trivial assumption to make. In the last analysis, however, even among these very modest assumptions there is none which I wish to treat as a foundational axiom. My contention is, rather, that none of the results I derive from them is so implausible as to force us to modify or surrender them. 25 We should not express the point by saying that the beliefs and desires provide people with their motivating reasons, for the reasons which motivate are the same reasons which the actor recognizes as normatively valid, hence they are facts, not mental states. (I owe this point to Jonathan Dancy.) However, the action is causally explained by mental states, including the belief of the actor that these facts provide her with valid reasons. 26 The ceteris paribus clause covers the case in which, besides desires with a maximizing structure, you have desires with a binary structure as well.

10 10 CHAPTER Presupposing Strategic Rationality My use of game theoretical concepts will be extremely modest. Nowhere the argument will rely on any complex mathematical result. The point of using these concepts is only to make clear the structural differences between situations of interdependent outcomes ("games") which may help or hinder the development of patterns of interdependent choices, and, most importantly, may help to explain why in some cases particular assumptions are needed, especially motivational ones, which are not needed in others. I also want to show how in each of these situations, patterns of interdependent choices may be predicted to maintain themselves once they have somehow arisen. However, I will not be much concerned with the vexed question of the emergence of norms. One reason is that no answer to this question is necessary for addressing the problems of legal philosophy I am interested in. All I need is an account of what it means for a social rule to exist. It is true that in the conventionalist theories which have been developed so far, the questions of the emergence and the maintenance of social rules have been treated as very closely related or even identical. But on my account these questions are to be kept apart. At this point I cannot fully explain the reasons for this insistence 27, but there is one pertinent observation I should make. The game-theoretical notions I will use are those of classical game theory. The very notion of interdependent reasons presupposes agents who are aware of each other's capacities of practical reason. Their form of reasoning is strategic: for every option they consider, they take into account that if the option is a reasonable choice, the others will anticipate it being made. This assumption is characteristic for what I call "classical" game theory: it makes a rather strong assumption of common knowledge. The players know the options each of them has, the outcomes of each combination of choices and the utilities each of them attributes to each of these outcomes: in short they know the game they are playing. They also know that the others are "perfectly" rational. Not only do they have this knowledge, they ascribe it to each other, know that the others ascribe it to them etc. In the interest of realism I will at certain points relax these assumptions of perfect common knowledge, introducing some uncertainty concerning the game (in particular the relevant utilities of other players) and some fallibility in reasoning. But even then I will conceive of the people following any social rule as strategic agents. This is worth stressing because on the whole, the game-theoretical study of social norms in recent years has gone off in an altogether different direction, deviating from Lewis in this respect. The basic idea is that in an environment in which several "strategies" are played, and in which it is possible at any moment to try out every other possible strategy, the number of agents playing a particular strategy will reflect the success of the strategy in the recent past. This idea can be interpreted in terms of biological evolution. In that case, the measure of "success" is the survival of offspring, and the trying out of 27 See ch. 2.6.

11 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 11 new strategies is a consequence of the recombination of genes and of random mutation. Or the idea can be interpreted in terms of social evolution. In that case, it is assumed that people are able to identify each other's strategies and to measure the utility-return of those strategies, and tend to copy the most successful ones. However, they do not anticipate that others will second-guess their learning behaviour. 28 These agents all commit the parametric fallacy: each of them takes himself to be the only deliberating agent around, and considers the behaviour of the others as a constant environment of his choice, a fact of nature. This biological approach to the development of cooperative behaviour has been enormously fruitful. However, it explains at most the emergence of a convergent pattern of private rules, not the following of social rules. Now it is undeniable that human social relations are characterized by the existence of some form of common knowledge, even if this knowledge is not always perfect, i.e. completely "transparent". So for any pattern of behaviour which the evolution theories predict to emerge, we can ask: suppose the agents have common knowledge of this pattern, and suppose each of them has the capacity of reflective reasoning would that make a difference? Sometimes it will not. In these cases I am prepared to consider the possibility that the social norm is nothing but a crystallization of the underlying pattern of privaterule following behaviour. Given that pattern, common knowledge will follow naturally for people who have the capacity for it. 29 Even if the social rule emerged among strategic agents, it may be somehow significant that the behaviour it prescribes could have been produced by an evolutionary process as well. Sometimes, however, a behaviour pattern can be predicted to arise among non-strategic agents, when it cannot be predicted that it will be maintained by strategic ones, at least not if the other motivational assumptions are kept equal. It is even possible that the non-strategic agents can be exploited by the strategic agent who is aware of their non-strategic reasoning. 30 In those cases, it is clear that the evolutionary approach does not really explain the emergence of norms. This exemplifies its general weakness: it abstracts from potentially relevant information which human subjects obviously have at their disposal. 31 Even when both clever people and stupid ones exhibit the same patterns of behaviour, we cannot be sure that one explanation accounts for both patterns. 28 Sugden 1986; Skyrms 1996, both building on Maynard Smith Young 1996 assumes that people choose a best reply to the behaviour of others which they expect on account of recent sample evidence, cf. Schotter 1981, Kliemt This is still a form of parametric learning behaviour. For all but the most simple behaviour patterns, it is incoherent to assume that people will be able to identify them without recognizing that they exemplify the same kind of intelligent behaviour as their own "replying" choices. This is evident in Schotter and Kliemt, who require their agents to be able to identify very complex strategies. Young's subjects, on the other hand, start by making random choices and then wrongly ascribe meaning to other people's similar random choices, cf. also Bicchieri 1993, ch. 6; Skyrms 1996, ch E.g. if a pattern of convergent private rules has emerged in an iterative Coordination game, the pattern will only be reinforced by the development of common knowledge of those private rules. 30 One example is the agent who has adopted the unconditionally cooperative strategy in one-shot PD-games because he acts in an environment in which his chances to meet a similarly motivated agent are very high, cf. Skyrms 1996, ch Cf. Cudd 1993, 128; Sen 1998, x.

12 12 CHAPTER 1 For example, population dynamics can explain the development of the signalling systems which are used by certain species of birds and insects, 32 but it is obvious that this is a far cry from the linguistic competence of human beings. 33 One option which is only available to beings capable of strategic thinking, is to make variations on established communication patterns which are immediately recognizable to others. Common knowledge introduces a flexibility to conventions which is totally lacking in its evolutionary predecessors. What I intend to show is that, on the assumption of strategic agency, we may still be able to explain the maintenance of social rules, even if not their emergence Coordination Problems According to David Lewis, conventions tend to emerge and to be maintained among a number of persons who recurrently face a certain coordination problem. What kind of problem is this? Suppose we are parachutists dropped behind enemy lines. It is imperative that we join forces as soon as possible. So we want to go wherever we may expect the others to go. But this is true of everyone of us. No-one has any reason to go to any particular place. More generally, people have a coordination problem if the following applies to them: person P must choose from a number of alternative options a-1, b-1, c-1, and so on; R must choose from a-2, b-2, c-2, etc.; S must choose from a-3, etc., and each has a decisive reason to choose option a if and only if all the others do so too, and similarly for option b. The same, however, need not apply to the choice of c: a couple of meeting points is enough to make the choice problematic. It is not even necessary that everyone has the same choice of alternatives. For the classification of options as a or b, the only criterion is whether meeting points are generated, not whether these actions come under the same description in any other way. Of two alchemists, who think they can secretly make gold out of the ingredients O and H, one must bring O and the other H, and not both O or both H. It is not uniformity that is required but collaboration. 34 All those involved reason as follows: I must go where the others go, I've got to find out where that is. How will each of the others reason? "I must go where the others go, I've got to find out where that is. How will each of the others reason?" 'I must go,... and so on' and so on", and so on in an unending spiral. To determine what I must do, I must already know what I will do: for that fact is needed by the others to determine what they will do, and I need that fact to determine what I will do. There is no-one who, independently of the others, can make a justified choice and thereby provide grounds for the others to choose. Everyone waits for everyone else. Assume there is ground for the supposition that R chooses a: then P has a good reason to do the same. Everyone involved who can surmise the supposition of P, will 32 Skyrms 1996, ch Bennett 1964, ch Lewis 1969, 12.

13 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 13 expect P to choose a and thereby borrow a reason to choose a as well. (Even though they may simultaneously think P 's supposition unjustified). R may also belong to those involved. Their patterns of thought can again be replicated by a subsequent group. And so on. Each member of a subsequent, replicating group who was already a member of a previous group, gets an extra reason to choose a. In this way a cumulative series of replications develops which must overcome all indecision. But this development must have an independent point of departure; otherwise everyone remains imprisoned in the reciprocal interdependency of their expectations. Let us try to characterize games of (pure) coordination more precisely. What I have called 'meeting points' Lewis calls 'proper equilibria'. A Nash equilibrium is reached when neither chooser would have been able to achieve a better result if he had chosen differently on his own. The concept of a 'proper equilibrium' is a bit stronger. A proper equilibrium is achieved if it is true for each player that, given the other player s choice, he will not only fail to improve his outcome by making any deviant choice, but can only worsen it. 35 We initially define a Coordination game as any game with at least two proper equilibria. 36 Suppose, however, there are two proper equilibria, but both players vastly prefer one of those. In that case the situation intuitively does not seem problematic at all: each of us will head straight for the better outcome, confident that the other will do so as well. It is difficult to explain the rationality of this choice in the usual terms of rationalchoice theory. 37 It is sometimes suggested that this is a simple case of calculating the average expected utility for each choice. But this criterion cannot be used because, for example, for every alternative open to C, R would have to guess correctly the probability of C choosing that alternative. But R cannot determine how probable C's choices will be without an idea of the probabilities that C attributes to R's choices. And to know that, he would have to know what idea C had of the probabilities attributed to his choices by R, and so on. Expected utility reasoning in this type of context commits the parametric fallacy: you treat your co-player's choices as constants and your own as variables. 38 Throughout this book, as I announced in the previous section, I will assume people's reasoning to be strategic: they know the others to be equally rational as they are themselves. Then they also know that if a big if there is a determinate solution to their problem of choice, this choice will be anticipated by their co-player, and that 35 If we only require two equilibria, zero-sum games qualify as games of coordination. 36 Lewis actually requires two proper coordination equilibria (i.e. by unilateral deviance you can neither improve your own position nor that of your partner), but this is an unnecessary restriction, Den Hartogh 1985, ch Gibbard 1965; Regan 1980, 18ff; Sugden 1993, 73 ff; Hollis & Sugden One might suggest that in such situations only the best possible and the worst possible outcome are relevant: if for all choices the worst possible outcome is the same, one should go for the option with the best possible outcome. An alternative explanation is that in such situations a basic convention exists to "think as a team". See ch. 2, note Elster 1978, For this reason the founders of game theory, Von Neumann and Morgenstern, rightly did not allow probabilities into the domain of game theory, and to that extent the attempts of e.g. Harsanyi & Selten 1988 or Skyrms 1990 to develop game theory as a part of Bayesian decision theory are misguided. Cf. Mariotti 1996 for more technical criticism. However, if the assumption of common knowledge of rationality is given up, forms of game theory can be developed which rely on inductive evidence concerning other people's behaviour, see ch and note 28.

14 14 CHAPTER 1 should not be a reason to choose differently. 39 In other words: if any game has a solution, it must be an equilibrium. It does not follow that if any game has at least one equilibrium outcome, it must have a solution. 40 Whatever the explanation, it is clear that in a game with at least two proper equilibria it is only difficult to choose if the players have no common preference for one of those. In such a game, therefore, a coordination problem only exists if there are at least two proper equilibria which are equally valued by all players, provided there is no proper equilibrium which is valued higher. The equilibria must, at the same time, be optima Conventions Coordination problems are not always "solved", but solutions do occur. In explaining how this works, Lewis follows Thomas Schelling. 42 The basic idea is this: of the equivalent alternative meeting-points, choose the alternative that somehow stands out. You may assume that your co-player will follow the same maxim, so you may expect him to make this salient choice as well. That is a further reason to make it yourself. And you can assume that he realizes both that you will make that choice and anticipate him making it, and so on and so on. In this way, all parties take turns playing into one another's thoughts until the desired goal has been harmoniously reached. Reasoning in this way, "we are windowless monads doing our best to mirror each other, mirror each other mirroring each other and so on." 43 Lewis builds his theory of conventions on these ideas in the following way. Suppose that a certain coordination problem regularly reoccurs in a group, and assume that the first time the rendezvous succeeded possibly because all participants identified the same salient alternative, or perhaps because by chance all made the same choice. If the problem arises again, the situation is different in one significant aspect: amongst all acceptable alternatives, one stands out because it has already been chosen once before. That precedent provides the alternative with a particular prominence. Everyone has reason to choose that alternative in the expectation that everyone else has the same expectation too, etc. The next time the problem comes up, the same reasoning can be followed, this time with even greater confidence. Choosing this alternative emerges as a regularity and this regularity itself is reason enough to repeat the choice, 39 Luce & Raiffa 1957, 63-65, cf. Von Neumann & Morgenstern 1944, 48. Expected-utility reasoning does not satisfy the postulate, for if I ascribe any probability under 100% to your choice, and this leads me to a solution, you will anticipate this, and so your reaction should be predictable with 100% probability, which falsifies the assumption. 40 See ch. 2.3 & Lewis does not make this requirement. In ch we will find that an additional requirement should be made: there is no risk involved in doing one's part in converging on any of the meeting-points (or on minimally two of them); it is compatible with making a maximin-choice. 42 Schelling 1960, ch Lewis 1969, 38.

15 A CONVENTIONALIST THEORY OF SOCIAL RULES & OBLIGATIONS 15 again and again, so that it becomes self-confirming. A convention, Lewis concludes, is a regularity of conduct which reconfirms itself. 44 It is characteristic of a convention that everyone (usually) has convincing reasons to uphold it but at the same time everyone would also have had equally good reasons to uphold an alternative convention if that happened to have been the prevailing one. (Which side of the road we drive, which day dustbins are emptied or markets held, which part of town the shoe shops or night-clubs are concentrated, which paper we consult for housing ads, what image skull or what not is used to warn against poison, etc.). All possible conventions are equally good; which one becomes established is a matter of chance. In that sense, the prevailing convention is arbitrary. It only has the (incomparable) advantage of being established in fact. (Hence it is mistaken to conclude from the fact that some norms are "merely conventional" that there can be no compelling reason to adhere to them.) 45 Now Lewis' insight is that this arbitrary character of conventions can be related to the identifying feature of coordination problems: access to more than one equivalent proper equilibrium. The pub The Saracen's Head does not distinguish itself particularly favourably from the other pubs in town; any other pub would offer an equally pleasant ambience in which to meet one's friends. Yet we all go to The Saracen's Head because we all know it is the only place where we can be sure to meet one another Conventions in Assurance games There is an almost complete isomorphy between the philosophy of money and the philosophy of law. The first attempts to answer the question: why are people prepared to accept things without any intrinsic value, like pieces of paper, in exchange for nutritious loaves and beautiful pictures? The second asks: why do people accept legislative and judicial pronunciations as binding, whether they agree to their content or not? We find the same type of answer to both questions. The first school of thought in monetary philosophy is the theory of Natural Money. Gold and silver (or cattle, beads, rice, salt) have their own objective and permanent value, and therefore it is reasonable to accept and to issue pieces of gold and silver as tender for goods of equal value. If the circulation were to stop suddenly, the people left behind with the gold would not be the worse for that; therefore nobody needs to be afraid to take it. How about paper money? The Nederlandsche Bank who issues it assures me in print that on request it will supply the bearer a specified amount of those objective valuables. We know the Bank is capable of making good its promise, because the total amount specified by the paper issued is "covered" by the gold and silver in its possession. Therefore we only need to trust the Bank in order to take its paper. The value of this fiduciary money remains dependent on the objective value of the treasures stored in the Bank's vaults. 44 "A metastable, self-perpetuating system of preferences, expectations, and actions, capable of persisting indefinitely," Lewis 1969, E.g. Aristophanes, Clouds 1399ff, discussed in Nussbaum 1990, 222.