DRAFT VERSION

     How would one explain religion? I find most past attempts unsatisfactory, and I will try to show you today that we can now do better. This is not (or not just) hubris on my part; at any rate it is (emphatically) not a self-aggrandising claim, for I am not saying that I can explain it better, but that scientific developments for which I cannot claim any credit can help us finally understand, or understand much better, why there is religion and why it is the way it is. Yes, scientific progress means we have a much better grasp of why people have religious notions and norms than we would have had fifty years ago, and we are only at the beginning of this voyage of discovery. This is mainly because scientific explanations of how minds work have got incredibly better in the last few decades.

     I said that we could better understand why there is religion and why it is the way it is. I must emphasise this last point, for this is a major flaw in many theories about religion, from anthropology or philosophy (und leider auch Theologie, of course); they try to explain some ideal religion, or some local religion, or some reasonable reconstruction of religion, but do not take the full measure of religious concepts and norms as a culturally variable phenomenon. To be a perfectly obnoxious anthropologist for a very short time, let me remind everyone of facts we all know but should keep at the forefront of our attention when we discuss religion:

·    Most religious systems in the world are not about an eternal Creator

·    In fact the creation of the universe is of limited interest to most people in the world

·    People can have many gods, or a few gods, or a combination of several gods and many spirits, or a few spirits and many gods, or many ancestors and no spirits, etc.

·    Gods are said to die in many traditions

·    They are remarkably stupid in many others

·    The 'salvation of the soul' is alien to most people's ideas about death.

     I could carry on in this vein for quite some time, but then I would be the truly obnoxious anthropologist and that would serve no good purpose. The point of this exercise in diversity-awareness, as it were, is to emphasise how parochial, as it were, many accounts of religion can be. Religion probably does not stem from a desire to explain the origin of the universe, since most people get by perfectly well without any creation account; there is no instinct for transcendence in human beings, since the most frequent religious beings are ancestors who are assumed to be as real as the living, only more elusive; you cannot explain religion as moral coercion combined with promised rewards, since in many places the soul needs no salvation and will in due time become an ancestor.

     Paying attention to the true diversity of religious concepts and norms is certainly necessary, but it is far from sufficient. We can compile lists of different religious concepts and measure the relative frequency of particular notions. This is what anthropologists have done and this is a necessary starting point. But is it enough? To take a distant example, philologists have for a long time documented the variety of languages, the relative distance between them, their plausible historical connections, as well as established a catalogue of extant grammatical systems. But at some point linguists decided to explain linguistic structure, which in effect meant this: underneath the luxuriant variety of systems, there are a few underlying rules. These rules do not come from nowhere: they are the consequence of how human brains function.

     A similar scenario is conceivable for the diversity and underlying common features of religious ideas. As in the case of language, it implies that we should consider, beyond the actual concepts and norms that we call religion, the mental systems that support them. This I think is now possible, in a way that is quite different from what it would have been thirty years ago, because of our constantly increasing knowledge of the mind-brain.

     In the following pages I use various kinds of evidence to suggest how different mental systems are involved in the selection of religious concepts. The human mind is not a single system designed to produce an accurate representation of the world. Rather, it consists of multiple systems geared to representing and predicting various parts of the environment, or guiding action in different domains according to different principles. None of these evolved systems is about religion. But some of them may be activated, in the context of representing religious agents, in such a way that concepts of such agents have a high probability of transmission. So examining which systems are activated in this way, and how they fashion different aspects of viable cultural concepts, should explain not just why we have these religious notions but also why we do not have others, in other words explain the recurrent features of such concepts.

     This requires that we go beyond what people know and believe, to the underlying systems that support such knowledge and belief.    The main strategy in the study of religion so far was to just ask people about their religious concepts. This is of course an indispensable first step, but we cannot stop there. It is not just that people's explanations may be vague and idiosyncratic (though they are). It is also that we have no good reason to assume that people have much access to the cognitive machinery that produces those concepts. People after all have no access to the way their brains turn two-dimensional retinal images into three-dimensional visual representations, or to how they produce syntactic sentences. People can feel the difference between two sentences ('who did you see me with?' and 'who did you see me and?', respectively) without being able to explain why one is ungrammatical. The same point applies to concepts. Some notions are easier to acquire than others, some conceptual associations are better recalled, and some create stronger emotional effects. All this depends on processes largely beyond conscious access, in the same way as the workings of the visual cortex.

     What I am offering here is a multiple system explanation of religion. I do not believe in 'magic bullet', single-cause explanations of religion, not just because religion is complex, subtle, etc., but also because religion is a cultural phenomenon. It is something you get from other people and something you will contribute to transmit to others. What we call 'religion' are successful religious concepts and norms. That is, they are the ones that survived many cycles of individual acquisition and transmission. The rest, these possible variants that were entertained but then forgotten, or adopted by a few but distorted by others, these unsuccessful variants just do not register. That religion is successful religion, in this sense, suggests that it activates many different mental systems in ways that favour retention and transmission. The emphasis should be on multiple mental systems. A durably successful cultural institution is like a durably successful economy, which probably owes its perennial success not to one single factor (good natural resources, just enough people, the right kind of culture, a lucky history, etc.) but to the fortunate combination of most of these. The same goes for cultural transmission, so that religion is successful for many reasons instead of one. It may be frustrating for those who hoped that a single-shot account would do most of the explanatory work. As we will see, the brain-based explanation is more complex but also much more interesting than that.

First point: religious notions are supernatural notions

     The world over, people's supernatural repertoire includes a variety of concepts of imagined artefacts, animals, persons and plants: concepts of floating islands, of mountains that digest food or have blood circulation, of trees that listen, of animals that change species or of people who can disappear at will. These are found in folktales, anecdotes, myths, dreams and religious ritual and correspond to a small 'catalogue' of templates for supernatural concepts (Boyer, 1994a; 2000). We also find that a particular sub-set of these concepts is associated with more serious commitment, strong emotions, important rituals and/or moral understandings. An association between a supernatural concept and one or several of these social effects is our main intuitive criterion for what is 'religious' (Saler, 1993).

     Religious concepts are supernatural concepts, that they gain minimal salience by including a particular kind of expectation-violation. There are, to simplify matters a great deal, two major levels of conceptual information in semantic memory. One is that of kind-concepts, notions like 'table' and 'tiger' and 'tarmac' and 'tree'. The other consists of domain-concepts, such as 'intentional agent', 'man-made object', 'living thing'. Most of the information associated with these broader concepts comes in the format, not of declared statements (e.g., 'living things grow with age') but of intuitive expectations and inferences. Without being aware of it, one expects living things to grow, intentional agents to have goals and their behaviour to be caused by those goals, the structure of artefacts to be explained by a function and the latter by a designer's intention.

     Now supernatural concepts describe minimal violations of such expectations: a tree is said to listen to people's conversations, a statue is said to bleed on particular occasions, a person is described as being in several places at once, another one as going through walls, etc. Note that such concepts violate domain-level and not kind-level expectations. A talking tree goes against expectations not because trees in particular are usually silent but because plants are assumed to be non-intentional. Also, note that the violations are minimal, keeping in place all the (non-violated) default assumptions that usually accompany a given domain concept. A talking tree is still assumed to grow like all plants, ghosts that go though walls still perceive and represent their environment like other intentional agents. Indeed, these non-violated assumptions provide an indispensable grounding for people's inferences about supernatural entities and agents (Barrett & Keil, 1996; Boyer, 1994b).

     This twofold condition: [a] include a violation of domain-level intuitions and [b] allow inferences from relevant non-violated assumptions, is sufficient to account for the recurrent features of supernatural concepts the world over. That is, the subject-matter of fantastic imagination, dreams, folk-tales and religion generally revolves around a small catalogue of concepts built in that way (Boyer, 1994b). The concepts may be very different from one pace to another, but the templates are few, consisting of a combination of one particular domain-concept and one particular violation (e.g. 'intentional agent' and 'physical solidity' for the 'ghost' concept). Also, experimental work in different cultures suggests that concepts built in this way are more likely to be recalled than either predictable conceptual associations, or oddities constructed by violating kind-level associations. A table made of sausages (violation of kind-level expectations) may be quite striking, but in the end is not quite as easily acquired and recalled as a table that understands conversations (violation of domain-level expectations). This effect seems to work in fairly similar ways in different cultural environments (Barrett, 1998; Boyer & Ramble, 2001).

     Religious concepts are a sub-set of supernatural notions, with special additional features. But it is worth insisting on the fact that they belong to this broader domain, as this explains their mode of acquisition. In supernatural concepts, most of the relevant information associated with a particular notion is given by domain-level intuitions. In other words, it is spontaneously assumed to be true in the absence of contrary information. This is why no-one in the world needs to be told that ghosts see what happens when it happens, or that gods who want some result will try to do what it takes to achieve it: such inferences are given for free by our specialised mental systems (intuitive psychology in this case). In religion as in other supernatural domains, the violations are made clear to people, but the rest is inferred. Concepts that are both salient (because of the violation) and very cheaply transmitted (because of spontaneous inferences) are optimal from the viewpoint of cultural transmission.

     Now some supernatural concepts matter much more than others. Whether Puss-in-boots did run faster than the wind or not is of no great moment, but whether the ancestors did find out that someone sacrificed to them certainly is. The question is, why do some concepts of imagined entities and agents rather than others matter to people? Because, I will argue, other specialised mental systems are involved in their representation. In the following pages I will outline the ways in which this occurs, that is, how religious concepts are associated with intuitions about agency, about social interaction, about moral understandings and about dead bodies.

Religious concepts are about agents

     Although there are many templates for supernatural concepts, the ones that really matter to people are invariably person-like. There is certainly a tendency in the human imagination to project human-like and person-like features onto non-human or non-person-like aspects of the environment; such representations are attention-grabbing or enjoyable; they are certainly found in many aspects of religious agency, as Stewart Guthrie has demonstrated (see Guthrie, 1993). As Guthrie points out, such projections do not stem from an urge to make various situations or occurrences more familiar or more reassuring (which is seldom the result anyway) but to afford richer inferences about them. Projections of human-like features add complexity to the world, which is why they are easily created and transmitted by human minds (Guthrie, 1993).

     This constant search for relevant inferences may well be the reason why the 'anthropomorphism' of religious concepts is in fact rather selective. That is, the domain of intuitions and inferences that is projected is intentional agency, more frequently and more consistently than any other domain of human characteristics. Besides, intuitions concerning intentional agency are activated not just when interacting with humans, but also in our dealings with animals (Boyer, 1996). This is why one can postulate an intentional agent around, and run various inferences about what it can perceive, what its next reactions might be, etc., without making it a human person in other respects (Barrett, 2000).

     This is consistent with developmental and other cognitive evidence concerning the complex intentional psychology or 'theory of mind' present in all normal human minds. This 'mind-reading' system is geared to interpreting other agents' (or one's own) behaviour, as well as figuring out what their goals, beliefs, intentions, memories and inferences are. Rudimentary forms of such mind-reading capacities appear very early in development (see for instance Meltzoff, 1999), they develop in fairly similar forms in normal children. Their working is out of reach for conscious inspection; only the outcome of their computations is conscious.

     There are two distinct 'origin scenarios' for our capacity to understand intentional agency, to create representations of other agents' behaviour, beliefs and intentions. They could be called the social interaction and predation explanations respectively.

     A widely accepted evolutionary scenario is that we (higher primates) evolved more and more complex intentional psychology systems to deal with social interaction. Having larger groups, more stable interaction, and more efficient co-ordination with other agents all bring out significant adaptive benefits for the individual. But they all require finer and finer grained descriptions of others' mental states and behaviour. This is why we find, early developed in most humans, a hypertrophied 'theory of mind' that tracks the objects of other people's attention, computes their states of minds, predicts their behaviour (Whiten, 1991 Povinelli & Preuss, 1995 Meltzoff, 1999).

     Another possible account is that at least some aspects of our Theory of Mind capacities evolved in the context of predator-prey interaction (Barrett, 1999). A heightened capacity to remain undetected by either predator or prey, as well as a better sense of how these other animals detect us, are of obvious adaptive significance. They are present in primates and especially in humans, who made up for an unimpressive physique by being just better at figuring out what other animals will do next, hence our record as efficient hunters. In particular, hunting and predator-avoidance become much better when they are more flexible, that is, informed by details about the situation at hand, so that one does not react to all predators or prey in the same way. Indeed, some primatologists have speculated that joint detection of predators may have been the primary context for the evolution of agency-concepts (Van Schaik, Van Noordwijk, Warsono, & Sutriono, 1983). In the archaeological record, changes towards more flexible hunting patterns in modern Humans suggest a richer, more intentional representation of the hunted animal (Mithen, 1996).

     These accounts are not necessarily incompatible, since they may be relevant to different sub-systems involved in the representation of agency. So far, I have followed the classical, slightly oversimplified description of intuitive psychology as a single system dedicated to the intentional description of behaviour. But it is very likely that several, relatively autonomous systems are engaged in such performance. That is, there may be a specialised system detecting goals from motion, while another one only deals with emotional cues on the faces of conspecifics, another is busy representing other agents' beliefs, another detects their presence on the basis of unexpected noises, etc.. There is mounting experimental and clinical evidence for this fractionation of Theory of Mind. A disorder like autism stems from an inability to represent other people's thoughts, but it does not seem to impair primitive animacy-detection (realising that some objects in the environment are goal-directed) or gaze-following (Baron-Cohen, 1995). Williams' syndrome children are very good at detecting, following and displaying emotional cues relevant to social interaction, although they often have a very poor understanding of the beliefs and intentions that motivate behaviour (Tager-Flusberg & Sullivan, 2000). In a similar way, chimpanzees may pay attention to gaze-direction without associating it with specific intentions, which shows that these two capacities are separable (Povinelli & Eddy, 1996).

     Now, if there are several distinct Theory of Mind components, there may be different ways in which human minds can postulate agents without much evidence. Indeed, I would argue that supernatural agents are made salient and relevant to human minds by two distinct routes, each of which contributes a particular aspects of these imagined agent. The first route is through those systems we developed in predator-prey interaction; the second one is though those systems that are especially dedicated to social interaction.

     As Guthrie emphasised, detecting agents around, often on the basis of scant or unreliable evidence, is a hallmark of human minds. When we see branches moving in a tree or when we hear an unexpected sound behind us, we immediately infer that some agent (animal or human) is the cause of this perceptually salient event, and that some goal of that agent explains its behaviour. Note that the systems that detect agency do not need much solid evidence. On the contrary, they 'jump to conclusions', that is, give us the intuition that an agent is around, in many contexts where other interpretations (the wind pushed the foliage, a branch just fell off a tree) are equally plausible. There are many everyday situations where we detect agency and then abandon this interpretation, once we realise there was no agent around. But, that is the important point, we spontaneously create these interpretations anyway. For Justin Barrett, there are important evolutionary reasons why we (as well as other animals) should have 'Hyper-Active Agent Detection'. In a species evolved to deal with both predators and prey, the expense of false positives (seeing agents where there are none) is minimal, if we can abandon these misguided intuitions quickly. By contrast, the cost of not detecting agents when they are actually around (either predator or prey) could be very high. So our cognitive systems work on a 'better safe than sorry' principle that leads to hyper-sensitive agent detection (Barrett, 2000).

     According to this evolutionary interpretation, predation-related capacities not only makes it easy to detect agents when there is little or no evidence for their presence, but also informs some of their features. For one thing, our agent-detection systems trigger emotional arousal in a way that is quite automatic. That is, these systems lead us intuitively to assume, not just that there are agents around but that this presence may have rather dramatic consequences for us. This is a feature that directly translates in supernatural imagination. People may well imagine all sorts of supernatural agents that are irrelevant to their well-being (like elves, goblins and suchlike); but the ones whose traces that people think they saw are generally of great emotional import. Second, agents are postulated, not on the basis of direct perception of their presence, but of indirect cues. As Barrett points out, what people claim to perceive are more often 'traces in the grass' than 'faces in the clouds'. A sudden noise, an unexplained shadow, a broken twig or someone's sudden death are explained as indices of the spirits' presence; this is far more frequent than a direct encounter with those agents. This feature too makes much more sense once we understand the contribution of predation-related mental systems, which after all are by design concerned with the interpretation of indirect cues and fragmentary signals as evidence for some agent's presence.

     This evolutionary account makes good sense, although association with predation-related intuitions is probably only part of what makes religious concepts salient. In particular, we need additional factors to explain what makes people's notions of supernatural agents so stable and plausible, and why they are so strongly informed by what other people say. To take the first aspect, religious concepts are much less transient than experiences of 'hyper-active agent detection', that is, of interpreting some noise or movement as the presence of an agent. The latter are often discarded as mistakes. As I said above, it makes sense to 'over-detect' agents only if you can quickly discard false positives, otherwise you would spend all our time recoiled in fear, which is certainly not adaptive. But thoughts about gods and spirits are not like that. These are stable concepts, in the sense that people have them stored in memory, re-activate them periodically and assume that these agents are a permanent fixture in their environment. Now consider the sources of information that shape people's religious concepts. True, having experiences of elusive shadows and sounds probably strengthens the general notion that there may be unseen agents around. However, it is also striking that the details of such representations are generally derived, not from what one has experienced (hyper-active detection) but of what others have said. People take their information about the features of ghosts and spirits and gods, to an overwhelming extent, from socially transmitted information, not direct experience. Conversely, intrinsically vague experiences are seen through the conceptual lenses provided by what others said about the gods and spirits. To sum up, people know vastly more about gods and spirits by listening to other people than by encountering these mysterious agents.

     I insist on this seemingly obvious point because it introduces yet another way in which religious concepts activate mental systems. Information about gods and spirits mainly comes from other people. It is also connected to our representations of what other people believe and want in a crucial way. That is, the way people construe religious agents is informed by mental systems geared to describing and managing interaction with other human agents.

Religious concepts are about social interaction

     In another paper (Boyer, 2000) I emphasised a crucial difference between representations of other people and representations of possible supernatural agents. A good part of the information concerning a situation of social interaction is processed by specialised social mind systems. An important part of our mental architecture consists of inference systems that deal with social interaction. For instance, human beings are very good at:

monitoring social exchange, that is, finding out who is co-operating with whom, under what circumstances, as well as punishing cheaters and avoiding people who fail to punish cheaters (Boyd & Richerson, 1992; Cosmides & Tooby, 1992)

representing other people's personality, especially in terms of reliability, on the basis of indirect but emotionally charged cues (Bacharach & Gambetta, 1999; Frank, 1988);

building and maintaining social hierarchies, based either explicitly on resources or on indirect, seemingly arbitrary criteria for dominance (Sidanius & Pratto, 1999);

building coalitions, that is, stable co-operation networks where benefits are shared, the cost of others defecting is high, and measures are taken to pre-empt it (Tooby & Cosmides, 1996);

gossiping, that is, taking pleasure at receiving or imparting information on adaptively significant domains (sex, resources, hierarchy) about and with other members of one's social network (Gambetta, 1994);

and many other such social interaction tasks. It is quite likely that such capacities are supported by a variety of functional systems, so that one does not so much have a 'social mind' as social mind capacities.

     People engaged in a particular situation of social interaction have a representation of that situation, associated with representations of all sorts of background details. A subset of that information is such that it activates social mind systems. That I drive a red car may be non-strategic information (for most people in most contexts) but also strategic (for instance in a social setting where that indexes wealth or arrogance of hipness). That you saw a particular person yesterday may be non-strategic for me, or become strategic information if I suspect that the two of you are in some coalition. So whether some piece of information is strategic or not is in the eye of the beholder. To say that information is strategic does not mean that it is important. That two of my colleagues had an affair ten years ago is not really of great importance to me, but it is strategic information in the precise sense used here. To sum up, then, given a particular situation, there will be in the minds of the participants some information that we call strategic because it is handled by their social mind systems.

     Now all standard social interaction, from a young age, is based on a principle of imperfect access: that is, the assumption that other people (and we ourselves) only have partial access to the strategic information pertinent to a particular situation. By contrast, supernatural agents seem to be implicitly construed as perfect access agents. A tacit assumption is that, given a situation x, and given some information about it that would be strategic, the supernatural agent has access to it. I must emphasise a few points that may be ambiguous in the above formulation:

     This assumption often remains tacit. You do not need to represent an explicit principle like 'the ancestors have access to what matters to our social mind systems' anymore than we need to represent a principle of the form 'objects that are bounced against a wall will bounce at an angle equal to the collision angle'.

     The assumption does not require that people represent what the strategic information in question amounts to. You can represent that 'if there is strategic information about this situation, the ancestors know it' without having any description of the strategic information in question. (In the same way, your inferences work on the assumption that there is something in giraffes that makes them grow differently from horses, and that it is innate in giraffes, without knowing or indeed having any representation of what this 'something' is.)

     To illustrate this and some further arguments about concepts of gods and spirits, let me make use of Roger Keesing's vivid description of Kwaio religion (Keesing, 1982). The Kwaio concept of spirit-ancestor (adalo) illustrates my contrast between contemplative, theological understandings and the more mundane business of representing religious agents in practical contexts. The Kwaio live in the Solomon Islands; most of their religious activities, as described by anthropologist Roger Keesing, involve interacting with ancestors, especially the spirits of deceased members of their own clans, as well as more dangerous wild spirits. Interaction with these adalo (the term denotes both wild spirits and ancestors) is a constant feature of Kwaio life. People frequently pray to the dead or give them sacrifices of pigs or simply talk to them. Also, people 'meet' the ancestors in dreams. Most people are particularly familiar with and fond of one particular adalo, generally the spirit of a close relative, and maintain frequent contact with that spirit.

     Now Kwaio people need not be told that spirits can perceive what happens, or that they can make a difference between their wishes and reality. People are just told that, for instance, 'the spirits are unhappy because we failed to sacrifice a pig for them'. To make sense of that utterance one must activate one's intuitive psychology inference systems. In the same way, no-one is ever told that 'gods (or spirits or ancestors) have access to whatever is strategic in any particular situation'. What is made explicit is most often a vague assumption that the spirits or the gods simply know more than we do. But it seems that people in fact assume something much more specific, namely that the gods and spirits have access to strategic information (as defined here) rather than information in general.

     Kwaio people's statements about their ancestors highlight this. At first sight, what they say would seem to confirm that ancestors simply know more: 'The adalo see the slightest small things. Nothing is hidden from the adalo. It would be hidden from us [living people, but not from them]' or again 'an adalo has unlimited vision'. But when people illustrate these statements, notice how they immediately move from 'agents who know more' to the much more specific 'agents who know more about what is strategic': 'An adalo has unlimited vision [...] something happens in secret and [the adalo] will see it.; [if] someone urinates, someone menstruates [NB, in improper places: doing this is an insult to the ancestors] and tries to hide it, [...] the adalo will see it'. (1)

     In other words, although you can say that the adalo in general see what humans cannot see, what first comes to mind is that they can detect behaviours that would have consequences for social interaction: someone who has polluted a particular place puts others in danger and should perform appropriate purification rites. Whether someone did violate these rules or not is clearly strategic information. When people represent possible violations, this activates their inference systems for social interaction. For them, it also goes without saying that it is that particular kind of information that the adalo have access to. It may be hidden to people (this is the 'imperfect access principle': people's access to strategic information is not guaranteed) but not to supernatural agents (they have full access). People are told that 'Someone urinates in a house; we humans cannot see it; but that makes the adalo very angry,' or some other statement of that kind. Interpreting such statements requires that the adalo (or whatever supernatural agent people in your group talk about) have access to strategic information.

     The same remark would apply to agents (like the Christian god) described by theologians and other religious specialists as omniscient. Most believers would readily assent to statements such as 'God knows everything' or 'God sees everything'. This, however, does not mean that the religious agents is literally assumed to represent every aspect of every situation in the world. In people's conversations and trains of thought concerning God, it would seem that such statements as 'I bought broccoli and God knows about it' are somehow less frequent and salient than thoughts like 'He lied to her and God knows it' or 'I did my best and God knows it'. In other words, if something counts as strategic information, in the precise sense used here, it is more easily and naturally included in thoughts about God's thoughts.

     In detailed experimental work, Justin Barrette had shown that people's explicit notions of an omniscient God are combined with an intuitive understanding of God as having a human-like mind Barrett & Keil, 1996. A person may (explicitly) declare that the gods see and hear everything yet show them an offering or say a prayer out loud. The assumption of full-access to strategic information is just another aspect of this discrepancy between official, and explicitly entertained, descriptions of supernatural agents, and the intuitive assumptions activated when thinking about them. Now why is the full-access expectation so important to interaction? I would claim that it changes the way one considers a situation, and this has effects in several domains. One of them is the construal of moral understandings, and the other one is the possible link between religious agents and misfortune.

Religious concepts are parasitic upon moral intuitions

     A consideration of cognitive process involved in representing religious agents can help us discard a widespread but misleading account of religious morality. In this account, people are for some reason convinced of the reality of some supernatural agents; these are described as particularly anxious that people should follow particular moral rules; so people follow these precepts, often against their inclination. Against this, it seems that moral intuitions and understandings develop in all human beings because of specialised, early developed mental capacities connected with social interaction. This in turn creates all sorts of intuitions about possible courses of action. The intuitions do not require concepts of supernatural agents, but if there are such concepts around, moral intuitions will be associated with them. In other words, religious concepts are in part parasitic upon moral understandings.

     To understand why this is the case, we must fist examine the various ways in which people establish a link between their supernatural concepts and their moral understandings. I will call these three connections the legislator, exemplar and interested party models.

In the 'gods as legislators' model, there are moral principles because the gods or ancestors themselves decided what these norms would be. Several world-religions include lists of prohibitions and prescriptions, of varying length, attributed to some direct communication from the supernatural legislature.

     In the 'exemplar' model, some supernatural agents provide a model to follow. Saints or holy people are both different enough from common folk that they approach an ideal and close enough so their behaviour can serve as a model. This is the way people conceive of individuals with supernatural qualities like Gautama, Muhammad or the many Christian and Muslim saints as well as the miraculous rabbis of Judaism.

     Supernatural agents are also represented as 'interested parties' in moral choices. This means that the gods or the ancestors are not indifferent to what people do, and this is why we must act in particular ways or refrain from certain courses of action. Interaction with the Kwaio ancestors is of this kind. But the 'interested parties' model is much more general than that. We find it in many 'world-religions' whether or not the theologians find it acceptable. Most Christians entertain this notion that every single one of their moral choices is relevant to their personal connection to God. That is, God not only gave laws and principles but also pays attention to what people do. For obvious reasons, the notion that supernatural agents are interested parties is generally associated with the idea that the gods or spirits are powerful and that it is within their capacities to inflict all sorts of calamities upon people ' or help them prosper ' depending on their behaviour.

     In people's actual reasoning about particular situations, in the practical business of judging people's behaviour and choosing a course of action, the 'interested party' model is largely dominant. As far as anthropologists know, people in most places conceive of some supernatural agents as having some interest in their decisions. This can take all sorts of forms. The Christians for instance consider that God expects some particular kinds of behaviour and will react to departures from the norm. People who interact with their ancestors, like the Kwaio, have a much less precise description of what the ancestors want, but it is part of their everyday concerns that the adalo are watching them. In either case, people do not really represent why the ancestors or gods would want to sanction people's behaviour. It is just assumed that they will.

     When I say that this way of thinking about morality is 'dominant' I simply mean that it is constantly activated and generally implicit. It is the most natural way people think of the connection between powerful agents and their own behaviour. The 'legislator' and 'model' representations both have their limits. Explicit moral codes are often too abstract to provide definite judgements on particular cases (this is why scholars often augment them with a tradition of commentary and exegesis); conversely, moral paragons provide examples that are too specific to be easily applied to different circumstances.

     More important, however, is the fact that there is something intuitive and natural in the notion of agents that are attentive and responsive to the way people behave. Indeed, this notion may be rooted in the way moral intuitions are developed from an early age. A conventional view would be that children acquire moral concepts by generalising and gradually abstracting from social conventions. In this view, children start by noticing correlation between specific courses of action (e.g. beating up one's sibling or being noisy during class) and sanctions. They then abstract general principles of right and wrong from these specific cases. Also, children are described as building moral feelings by internalising people's emotional reactions to their actions (Gibbs, 1991). Both models may have underestimated the child's intuitive access to specifically moral dimensions of actions. Indeed, psychologist Eliot Turiel showed that even pre-schoolers have a good intuitive understanding of the difference between social conventions and moral prescriptions (so that beating up people is wrong even if no-one told you so, while being noisy is wrong only if there was an injunction to keep still) (Turiel, 1983; Turiel, 1998). Also, children found it much easier to imagine a revision of major social conventions (e.g. a situation where boys wore skirts) than a revision of even minor moral principles (e.g. a situation where it would be all right to steal an eraser). Finally, children make a difference between moral principles and prudential rules (do not leave your notebook near the fire-place!). They justify both in terms of their consequences but assume that social consequences are specific to moral violations (Tisak & Turiel, 1984).

     So experimental studies show that there is an early-developed specific inference system, a specialised 'moral sense' underlying moral intuitions. Notions of morality are distinct from those used to evaluate other aspects of social interaction (this is why social conventions and moral imperatives are so easily distinguished). They provide an initial basis on which children can understand adult moral understandings. This capacity for entertaining abstract intuitions about the moral nature of courses of action (without, of course, being able to explicate them) was found also in children with various amounts of experience with other children (Siegal & Storey, 1985), in different cultures (Song, Smetana, & Kim, 1987), and even in children with exceptional experience of abuse or neglect (Smetana, Kelly, & Twentyman, 1984, Sanderson & Siegal, 1988).

     That children have early moral concepts does not mean that they have the same moral understandings as adults, far from it. First, young children have more difficulty in figuring out other agents' intentions and feelings; second, they do not have a rich repertoire of past episodes to draw from when representing the key features of a situation; third, they may not be aware of local parameters of social interaction (they may resist sharing their toys with a cousin, noticing that their parents are not giving their car to the cousin's parents). But what is important here is that, from an early age, [a] children's moral understandings are founded on the intuition that some courses of action are right and others are not, [b] this intuition stems from feelings that cannot be further explicated, and [c] it is assumed that a course of action is right or wrong in itself, regardless of who is considering it. All three assumptions are found in adults too, and form an intuitive basis for moral inferences. Obviously, they are also supplemented by explicit understandings of moral principles as well as (in some places) their connections with religious concepts.

     The main conclusion to draw from this research is that moral understandings, far from being dependent upon socially transmitted (e.g. religious) explications, appear before such concepts are intelligible, in the same way regardless of what religious concepts are entertained by adults around the child, indeed they develop regardless of whether there are any religious concepts in the child's cultural environment. But another aspect of these cognitive findings is also important. Early developed moral understandings may well provide a context in which concepts of supernatural agents become more salient.

     To the extent that people represent a situation in a way that triggers particular moral intuitions and feelings, they generally assume that these intuitions are feelings are true regardless of who is considering the situation. They also assume that the only way to disguise the true moral nature of an action is to mislead people about the action itself. If you want to exculpate yourself, you cannot argue that beating up one's sibling is right, but you can claim that what seemed to be a beating-up was something entirely different. You assume that, to the extent that people share your information (or information you hold true) about what happened, they probably share the same moral intuition about it, and therefore will be led to react to it in similar ways.

     In other words, the way our moral intuitions allows for an empty place-holder, for the position of 'some agent who has access to my information about the situation at hand'. The moral system itself does not provide any description of that agent, although we try and make other people become such agents by explaining situations to them. Now, as I said above, some supernatural agents are represented by default as having full-access to strategic information. That is, people represent a given situation, and represent some information about it that is relevant to social interaction, they assume that the supernatural agent also has that information. (Again, all this consists in tacit assumptions). This means that, in morally relevant situations, a concept of god or spirit or ancestor is very likely to be activated as the most relevant way to fill in the empty place-holder. After representing a particular behaviour as wrong, and feeling the wrongness of it, it seems quite natural to assume that some other agent with full access has similar feelings.

     Indeed, this connection may gain additional salience from the fact that the moral intuition system, like other inference systems in our minds, is not really open to conscious inspection. That is, we have moral intuitions without having much access to the computational operations whereby the system produced such intuitions. Associating concepts of full-access supernatural agents to moral intuitions may well provide a post-hoc rationalisation of the intuitions. So in a sense concepts of gods and spirits are made more relevant by the organisation of our moral thoughts, which themselves do not especially require any gods or spirits. What I mean by 'relevant' is that the concepts, once put in this moral context, are both easy to represent and generate many new inferences. For instance, most people feel some guilt when acting in a way which they suspect is immoral. That is, whatever their self-serving justifications, they may have the intuition that an agent with a full description of the situation would still classify it as wrong. Now thinking of this intuition as 'what the ancestors think of what I did' or 'how God feels about what I did' provides an easy way of representing what is otherwise extremely vague. That is, most of our moral intuitions are clear but their origin escapes us, because it lies in mental processing that we cannot consciously access. Seeing these intuitions as someone's viewpoint is a simpler way of understanding why we have these intuitions. But this requires the concept of an agent with full access to strategic information. These associations are not the origin of the moral feelings, but a convenient way of commenting upon them. In this sense, moral intuitions and feelings contribute to the relevance of some supernatural concepts. The latter, again, can be seen as parasitic upon intuitions that would be there, spirits and gods or not.

Religious concepts exploit our intuitions about misfortune

     It may seem that gods and spirits matter to people mainly because these supernatural agents are described as having special powers. The ancestors can make you sick or ruin your plantations, God sends people various plagues. On the positive side, gods and spirits are also represented as protectors, guarantors of good crops, social harmony, etc. But why are supernatural agents construed as having such causal powers? Everything would be quite simple if these descriptions of supernatural agents were some kind of 'cultural postulates' from which people would then deduce that these agents do matter in particular circumstances. However, what seems to be a cultural 'postulate' is always the product of cultural transmission. There must be some processes that turn these propositions into plausible or indeed self-evident descriptions of what happens in the world. So even if people took the powers of gods and spirits as axiomatic, we should explain why they do that and do not adopt other 'postulates'.

The notion that gods and spirits matter because of their powers does not just beg the question of why they are represented as having such powers. It also creates difficult puzzles. For instance, in many places the most powerful supernatural agents are not the ones that matter most. The Fang of Cameroon and Gabon, among whom I conducted anthropological fieldwork, have all these rituals and complex emotions associated with the possible presence of the ghosts-ancestors. Now the Fang also say that the world (meaning earth and sky and all living things) was created by a god called Mebeghe, vastly more powerful than either the living or the dead. His work was completed by another god, Nzame, who invented all cultural objects: tools, houses, etc., and taught people how to hunt, domesticate animals and raise crops. Neither of these gods seems to matter that much. That is, there are no cults or rituals specifically directed at Mebeghe or Nzame, although they are assumed to be around, and they are in fact very rarely mentioned. For a long time, this puzzled many travellers, anthropologists and of course missionaries. Many African people seemed to recognise a Creator in the same sense as the Biblical one, yet were remarkably indifferent to Him. We will see below the explanation for this apparent paradox.

     For the time being, let us just keep in mind that what matters is not so much the 'powers' of supernatural beings considered in the abstract, as those powers that are relevant to practical concerns. In particular, ancestors, gods and spirits are readily mentioned when people represent or try to explain salient situations of misfortune. Indeed, this connection is so frequent that, for many non-specialists, it seems to provide an easy explanation for the origin and persistence of religious concepts. People are afflicted with various calamities, they cannot explain the amoral nature of their destiny, so they imagine gods and spirits that pull the strings. This, like many other popular 'origin of religion' scenarios, points to a real association, but in my view fails to appreciate the complexity of the mental operations involved.

     We often assume that people want to understand what happens to them. This is where gods and spirits, however feeble as explanations, at least provide some measure of intelligibility. A weak explanation is better than no explanation. But why would people want to understand their own misfortune? Why drives their minds to seek an explanation? Again, this seems to have an obvious explanation. Minds are designed that way, because a mind that produces a richer understanding of what happens (especially bad things that happen) is certainly better equipped for survival.

     Accepting this, it remains that some aspects of the association between religious agents and misfortune may seem paradoxical. To see this, let me return to the Kwaio example. The ancestors are generally responsible for whatever happens in a village: 'Spirits, a child learns early, are beings that help and punish: the source of success, gratification, and security, and the cause of illness, death, and misfortune; makers and enforcers of rules that must at first seem arbitrary' (Keesing, 1982 p. 33). Good taro crops and prolific sows indicate that the ancestors are happy with the way the living behave. Illness and misfortune are generally an effect of the ancestors' anger. True, the Kwaio like most people in the world accept that some events 'just happen' and have no particular cause. Some illnesses may be interpreted as a straightforward weakening of the body with no special implications; the fact that some ailments are cured by Western medicine shows that they are in that category of mere mishaps. But in general any salient event, particularly any remarkable misfortune, is seen as the action of the adalo. As a Kwaio diviner tells Keesing: 'If we see that a child is sick ['] we divine and then we sacrifice a pig [to the adalo].' Divination is required to understand which spirit is angry and why. A diviner will take a set of knotted leaves and pull them to see which side breaks first, indicating either a positive answer or no answer to a particular question. In most cases, ancestors are unhappy either because people have broken rules about what is proper and what is abu (forbidden or dangerous ' from the Oceanian root tapu that gave us our 'taboo'). Ancestors, like humans, crave pork and demand frequent sacrifices of pigs (Keesing, 1982 p. 115). Interaction with the ancestors can be quite complex, because it is not always clear which ancestor is causing trouble: 'If it is not really that adalo [discovered in divination] that asked for a pig, in order that our pigs or taro grow well, then even though we sacrifice it, nothing will happen'. So people may go through several cycles of divination followed by sacrifice to reach a satisfactory arrangement with the ancestors.

     This case highlights some very common features of the association between misfortune and religious agents. Although people assume that the ancestors are involved in many occurrences (like bad crops, illnesses, death, etc.) they do not bother to represent in what way they bring about all these states of affairs. That is, people's reasoning, when thinking about such situations, is entirely centred on the reasons why an ancestor would want them to fall ill or have many children, certainly not on the causal process whereby they make it happen.

     This is also true of other kinds of supernatural notions that people commonly associate with misfortune. One of the most widespread explanations of mishaps and disorders, the world over, is in terms of witchcraft, the suspicion that some people (generally in the community) perform magical tricks to 'steal' other people's, health, good fortune or material goods. Concepts of witches are among the most widespread supernatural ones. In some places there are explicit accusations and the alleged witches must either prove their innocence or perform some special rituals to pay for their transgression. In most places the suspicion is a matter of gossip and rarely comes out in the open. You do not really need to have actual witches around to have very firm beliefs about the existence and powers of witches. Witchcraft is important because it seems to provide an 'explanation' for all sorts of events: many cases of illness or other misfortune are spontaneously interpreted as evidence for the witches' actions. Witchcraft beliefs are only one manifestation of a phenomenon that is found in many human groups, the interpretation of misfortune as a consequence of envy. For another such situation, consider the widespread beliefs in an 'evil eye', a spell cast by envious people against whoever enjoys some good fortune or natural advantage Pocock, 1973. Witchcraft and evil eye notions do not really belong to the domain of religion, but they show that, religious agents or not, there is a tendency to focus on the possible reasons for some agents to cause misfortune, rather than on the processes whereby they could do it.

     Now this is not exactly what you would expect from a mind that is equipped to seek explanations for misfortune, as a way to avoid further bad occurrences. Or rather, one would imagine that such a mind would at least consider the ways in which bad things happen, for they are at least as important to predicting and avoiding them than some agents' reasons for making then happen. Indeed, this is often the way we think of occurrences that are mostly within our control. We do not bother to represent for what reason people would want to steal our car, but we take precautions based on our notion of how people could manage to steal it. By contrast, for these occurrences that largely escape control, people rather focus on the reasons of an agent. The ancestors were angry, the gods demanded a sacrifice, or the god is just cruel and playful. But there is more to that. The way these reasons are expressed is, in a great majority of cases, supported by our social exchange intuitions.

     People focus on an agent's reasons for causing them harm, but note that these 'reasons' always have to do with people's interaction with the agents in question. People refused to follow God's orders; they polluted a house against the ancestors' prescriptions; they had more wealth or good fortune than their God-decreed fate allocated them; and so on. All this supports what anthropologists have been saying for a long time on the basis of evidence gathered in the most various cultural environments: Misfortune is generally interpreted in social terms. But this familiar conclusion implies that the evolved cognitive resources people bring to the understanding of interaction should be crucial to their construal of misfortune.

     To return to our examples: The Kwaio ancestors afflict people with some disease because they want some sacrifice. In some of these cases people admit that they should have performed the sacrifice to start with. They are guilty of 'neglecting' a particular ancestor. They failed to maintain proper relations with him. These are clearly construed as exchange relations. Ancestors provide some form of protection and people provide roasted pigs. In some cases people tend to think that the ancestors are 'pushing it' a little and feel justifiably resentful. This is the kind of emotion that we find in situations where one party in social exchange seems to be increasing benefits without paying increased costs. In other words, relations with ancestors are framed by understandings and associated emotions that are intuitively applied to social exchange. Witchcraft, too, seems to be clearly construed as unfair exchange. The witches are trying to reap some benefit without paying any cost: witches are quite clearly 'cheaters' in the technical sense. Indeed, that is precisely what people like the Fang and many others who have witchcraft concepts say about witches: They are the ones who take but never give, who steal other people's health or happiness, who thrive only if others are deprived. Finally, the evil eye is a case people someone represent misfortune as caused by someone else who takes them for cheaters. People interpret your benefits as having cost you nothing. In their view, your benefit should be compensated by a cost. You think they do not perceive any cost in your good fortune, therefore they are jealous of you and this creates calamities.

     To sum up, then, people represent misfortune by using their social exchange templates. Evil spirits are enforcers of unfair deals, the angered ancestors are enforcers of fair ones, evil-eye people are over-reacting cheater-detectors and witches are genuine cheaters. These examples show that some prior mental process describes misfortune in such a way that it makes sense to include gods and spirits in an explanation of what caused it. That is, people's thoughts about such salient events are organised by the mental templates of social exchange. These do not necessarily require gods and spirits. But if you have concepts of gods and spirits, it is not too surprising that they should sometimes be included in explanations of salient events. If your representation of misfortune generally treats it as an effect of violations of social exchange, it will potentially include any agent with whom you interact. But spirits and gods are precisely represented as engaging in social interaction with people, especially in social exchange. So they are among the potential candidates for 'originators of misfortune', just like neighbours, relatives and envious partners.

  (1)    Keesing, 1982, p. 42.