The evolution of inferentially potent displays and their implications for cultural evolution

This paper lays out an evolutionary theory for the cognitive foundations and cultural evolution of the extravagant displays (e.g., ritual mutilation, animal sacrifice, and martyrdom) that have so often tantalized social scientists, as well as more mundane actions that influence cultural learning and historical processes. In Part I, I use the logic of natural selection to build a theory for how and why seemingly costly displays influence the cognitive processes associated with cultural learning—why do “actions speak louder than words.” The core idea is that cultural learners can avoid being manipulated by their potential models (those they are inclined to learn from) if they are biased toward models whose actions/displays would seem costly to the model if he held beliefs different from those he expresses verbally. I call these actions inferentially potent displays. Predictions are tested with experimental work from psychology. In Part II, I examine the implications for cultural evolution of this evolved bias in human cultural learning. The analytical model shows that this learning bias creates evolutionarily stable sets of interlocking beliefs and individually-costly practices. Part III explores how cultural evolution, driven by competition among groups stabilized at alternative sets of these interlocking belief-practice combinations, has led to the association of costly acts, often in the form of rituals, with deeper commitments to group beneficial ideologies, higher levels of larger-scale cooperation within groups, and greater success in competition with other social groups or institutions. Predictions are explored with cross-cultural, ethnographic, ethnohistorical and sociological data. I close by briefly sketching some further implications of these ideas for the study of religion, ritual, and costly signaling. Running head: Inferentially potent displays and cultural evolution Correspondence: 2136 West Mall, University of British Columbia, British Columbia, Canada, [email protected] Inferentially potent displays and cultural evolution 2 Researchers from across the social and biological sciences have long proposed a connection between seemingly costly displays, often in various ritualized forms—such as firewalking, ritual scarification, animal sacrifice, and subincision—and deep levels of commitment to group ideologies, religious beliefs, and shared values that promote solidarity and in-group cooperation (Atran and Norenzayan, 2004; Cronk, 1994; Durkheim, 1995; Irons, 1996; Radcliffe-Brown, 1952; Rappaport, 1999; Sosis and Alcorta, 2003a). This paper provides a novel approach to understanding these observations by considering how natural selection might have shaped our cognitive processes for cultural learning so as to give salience to certain kinds of displays or actions, and what the implications of such cognitive processes are for cultural evolution. Since the goal of this paper is merely to get this approach on the collective table, where it can properly compete with alternatives, I aim to provide a prima facie case for considering these ideas, and not a set of conclusive tests. The argument proceeds in three parts. Part I lays out a theory for the evolution of one particular component in the suite of cognitive adaptations that make up the human capacity for cultural learning. The core idea is that, with the evolution of substantial communicative capacities in the human lineage, cultural learners are potentially exploitable by manipulators who can convey one representation but actually believe something else. To contend with this adaptive challenge, I propose that learners have evolved to use, alongside the linguistic expressions of those individuals whom they are learning from (their models), inferentially potent displays by these models. Such displays or actions provide the learner with reliable cues that the model is actually committed to (i.e., believes in) the representations that he has expressed symbolically (e.g., verbally, written) and cheaply. Learners should use such displays in determining how much 1 Subincision is a common traditional ritual practiced in many societies throughout the world. It involves slitting the penis lengthwise along the urethra. As part of rites of passage, it is done to boys and adolescents without anesthetic. Inferentially potent displays and cultural evolution 3 to commit to a particular culturally-acquired mental representation (an ideology, value, belief, strategy, preference or practices). After laying out this idea, I summarize supporting experimental evidence from psychology. Building on the existence of this aspect of our evolved cultural learning capacities, Part II explores whether such a bias could create interlocking sets of beliefs and costly practices that are self-stabilizing. That is, can this adaptive learning bias lead to the emergence of stable combinations of widespread beliefs and costly practices (displays) in a social group, which could not otherwise persist (remain stable). My formal model reveals the conditions under which we expect to find interlocking sets of costly practices and beliefs. Such stable cultural evolutionary states are interesting because they show how particular displays or acts, which appear costly to one who does not hold the relevant corresponding belief, are sustained. Part III considers the possibility that if this evolved bias for using inferentially potent displays can stabilize interlocking set of beliefs and costly practices, it can also sustain costly practices that elevate the commitment of group members toward beliefs and practices that promote group benefits, larger-scale cooperation and solidarity, and in particular, favor success in competition with other social groups or institutions. This competition among stable equilibria culturally-evolved states favors social groups or institutions that are increasingly constituted by combinations of beliefs, that favor in-group cooperation/harmony and out-group competition, and practices (e.g. rituals) that maximize participants’ commitment to those beliefs. To explore the plausibility of these ideas in Part III, I summarize evidence from sociology and anthropology confirming three predictions from the model: (1) participation in costly rituals is associated with prosocial in-group behavior, because costly rituals transmit commitment to group beneficial beliefs/goals to participants, (2) institutions requiring costly Inferentially potent displays and cultural evolution 4 displays are favored by cultural evolution because costly displays by members transmit higher levels of belief-commitment and thereby promote cooperation and success in inter-group or inter-institution competition, and (3) costly rituals spread culturally from less successful groups to more successful groups because costly rituals affect groups’ success by increasing commitment to the group and galvanizing greater cooperation. In the concluding General Discussion, I consider some further implications of this line of reasoning, with a focus on religion, by explaining how such cultural evolutionary processes might help explain why (1) religions are often associated with prestigious paragons of virtue who make (or made) costly sacrifices, (2) martyrdom is powerful, (3) religions and rituals are loaded with sacrifices of various kinds, (4) counter-intuitive agents (e.g., gods or ancestors) want costly acts, (5) religious leaders take vows involving sex, fasting, and wealth, and (6) Mickey Mouse is not a god. I hope these preliminary suggestions may eventually complement the important ongoing work at the interface of religion, cognitive science, and evolution (Atran, 2002; Atran and Norenzayan, 2004; Boyer, 2001; Whitehouse, 2000). The Evolution of our Cultural Capacities In the last three decades the application of the logic of natural selection to the evolution of social learning, and in particular to the evolution of our species’ capacity for cultural transmission, has produced an array of novel theoretical insights, hypotheses, and empirical findings (for reviews see: Henrich and McElreath, 2006; Henrich and McElreath, 2003; Richerson and Boyd, 2005). One of the central lines of inquiry arising from this research program has focused on how our cultural learning processes can more effectively and efficiently acquire adaptive ideas, beliefs, values, preferences and practices from the others in our social world. The set of related hypotheses about these cognitive-operational details can be partitioned Inferentially potent displays and cultural evolution 5 into two categories, those based on contextual cues (e.g., based on the frequency of a belief or practice, or on model-based cues like prestige, success, ethnicity, or sex) and those derived from content cues (e.g., the details of the mental representations themselves that make them more memorable, more desirable, etc.). Below, I briefly review work in this area in preparation for laying out the inferentially potent displays hypothesis, the central evolutionary idea in this paper. Context biases Contextual mechanisms use cues to allow learners to more effectively extract and integrate information from the range of models available to them in their social milieu (Henrich and McElreath, 2003). Two specific cognitive mechanisms have been proposed, formally modeled, and empirically substantiated. The first, often glossed as prestige-biased transmission (Henrich and Gil-White, 2001), proposes that cultural learners uses model-based cues to figure out who, among their potential models (those from whom they could learn), is mostly likely to posses adaptive information (i.e., mental representations) suitable to the learner’s current situation (e.g., his/her role in the social group). Theory suggests, and a wide range of empirical findings have confirmed, that both children and adult preferentially pay attention to and learn from prestigious/successful models, especially when they match the learner on ethnicity (marked by dialect, dress, etc.) and sex (Henrich and McElreath, 2006; Henrich and Gil-White, 2001; Henrich and Henrich, 2007: Chapter 2). These effects influence cultural transmission across a wide range of representations, including opinions, economic decisions, food preferences, strategies, beliefs, technological adoptions and dialect. Moreover, these biases appear to operate across domains of specialty, as experts or stars in one field or endeavor (e.g., basketball) are granted influences in other arenas (e.g., clothing choice or politics). Anticipating what is to come below, if a highly prestigious individual was so inclined, he could express an opinion or Inferentially potent displays and cultural evolution 6 preference different from their own, that—once adopted by others—could yield benefits to him and costs to the learners. The second cognitive mechanism, termed conformist transmission, focuses on how learners can best weigh and integrate observations from multiple models (Henrich and Boyd, 1998). Learning mechanisms that ‘copy the majority’, ‘average what most prestigious individuals are doing’ or otherwise blend information from different models (apply a robust estimator: Boyd and Richerson, 1985: Chapter 7) allow learners to effectively aggregate information across models and reduce transmission noise (i.e., errors introduced during the process of observation and inference in learning). Such processes allows learners to extract mental representations that are more adaptive, on-average, than anything learners could acquire from a single model (Henrich and Boyd, 2002). As with the prestige-biased transmission, substantial amounts of empirical work from across the social sciences confirm the predictions derived from formal models of conformist transmission (Henrich and McElreath, 2006; Henrich and Henrich, 2007: Chapter 2). Content biases Evolutionary approaches to cultural transmission also provide a rich set of cognitivelyinformed hypotheses regarding how the content of representations influences their transmission (Boyd and Richerson, 1985: Chapter 5). The general insight here is that learners should pay particular attention to and remember representations likely to contain information that is more likely to be adaptive for the learner. Specifically, cultural learners should be more likely to pay attention to and recall representations when these are judged, ceteris paribus, more (1) potentially actionable, (2) emotionally evocative and (3) plausible or compatible: 1) Potentially actionable means that the content of a representation leads to inferences that can readily influence action, including additional inferences. Representations, for example, in Inferentially potent displays and cultural evolution 7 which the causes of unpleasant circumstances (e.g., storms or illnesses) are random with respect to the actions of those afflicted don’t lead to useful or helpful inferences or action, and thus are not easy to maintain. Evolutionarily non-actionable representations need not be stored because they can’t help you even if you do remember them. Instead, believing illness are caused by the jealously of others (e.g., the “evil eye”) can lead to inferences about who might be causing a particular illnesses and how to avoid such illness in the future. 2) Emotionally evocative responses to representational content yield a measure of potential fitness-relevance, at least in ancestral environments. Representational content that sparks more positive or negative emotional responses should be biased in storage, recall, and potentially in transmission. Evolutionary approaches regarding the origins of emotional or affective reactions to particular stimuli or content permit a wide range of more precisely specified hypotheses, see examples below. 3) Plausible or Compatible involves a variety of expectations that a learner might have about how the world works and, consequently, what is more and less likely to be true. Some expectations rely heavily on our reliably developing intuitions, including cognitive processes related to such domains as mechanics and biology. For example, representations from modern physics, which involve objects (e.g. electrons) that exist only probabilistically at any point in space, violate our intuitive expectations from folkmechanics and thus don’t readily transmit. Such plausible or compatible content biases can also be culturally acquired, such that the possession of one mental representation biases the acquisition of others. That is, having acquired a particular idea via cultural transmission, a learner may be more likely to acquire another idea, because the two “fit together” in some cognitive or Inferentially potent displays and cultural evolution 8 psychological sense. For example, believing that performing a certain ritual in the spring will increase the crop harvest in the summer might favor the acquisition of a belief that a similar ritual will increase a woman’s odds of conception, a healthy pregnancy, and the successful delivery of a robust infant. Hypotheses generated by an evolutionary approach to representational content biases have found a wide range of empirical support. For example, Barrett (2007) has shown that children from both the Ecuadorian Amazon and Los Angeles recall experimentally-transmitted information about the dangerousness of novel animals better than name-labels (e.g., peccary) or diet information (e.g., herbivore). Information about danger is emotionally evocative (fear), actionable (one can avoid these animals) and plausible (actual images of actual animals were used so teeth, size, etc. cues would not violate plausibility). Similarly, Fessler has proposed that since meat is more likely to carry pathogens and parasites dangerous to humans than other foods taboos directed at meat ought to be more prevalent than for other kinds of foods. Consistent with this, extensive analyses of cross-cultural data have shown that meat (including fish) are by far the most tabooed category of foods (Fessler, 2003). In this framework, the argument is that we are biased to be emotionally evoked by meat (disgusted) compared to other foods. Taboos on meat are also potentially actionable (avoid those foods) and plausible (some animals are toxic). Research on how the content of stories (e.g., rumors, urban legends) influence their differential cultural transmission shows that successful stories are both more plausible and more emotionally evocative than less successful variants. Heath and colleagues (2001) show how the differential success of urban legends is influenced by both their judged plausibility and their emotionally evocativeness (principally focusing on their disgustingness). Similarly, older lines of 2 Aspects of these heuristic categories for content biases are similar to parallel approaches such as Relevance Theory (Wilson and Sperber, 2004). Inferentially potent displays and cultural evolution 9 research on rumor demonstrate that content related to the three C’s, crisis, conflict, and catastrophe (evoked negative emotions) favor the success spread of representations (Rosnow, 1980). Exploring a content-bias hypothesis related to religious concepts, research has demonstrated how the presence of counterintuitive content in concepts or narratives biases longer-term memory (Barrett and Nyhof, 2001) in a manner that would favor such concepts or narrative in cultural evolution. Counterintuitive concepts or events violate our core assumption about the nature of things in the world, usually about intentional beings, animals, inanimate objects, or events (expectations often associated with folkphysics, folkpsychology, and folkbiology). Examples of counterintuitive concepts from this literature are “a person who can be in two places at once” (Boyer and Ramble, 2001) and a “thirsty door” (Norenzayan et al., 2006). The presence of a few counterintuitive concepts in a narrative, even within a list of otherwise ordinary concepts, improves long term memory for the entire narrative or list. From the perspective just presented, many counterintuitive concepts probably create complex mixtures of plausibility, applicability, and emotional evocativeness. Many religious beliefs, for example, would appear to be less plausible, more applicable, and more emotionally evocative than alternative non-religious concepts or explanations. Counterintuitive concepts, by their very nature, make stories or beings seem less plausible, which is likely why the optimal number of such violations is small. In an analysis comparing a sampling of successful and unsuccessful Grimm’s fairy tales, successful (wide known) fairly tales had 2 or 3 counterintuitive violations (Norenzayan et al., 2006). Counterintuitive properties are also like to generate emotional responses, like fear or interest (see Fredrickson, 1998 for a discussion of positive emotions like interest). Inferentially potent displays and cultural evolution 10 While her findings are far from conclusive, Mead’s study of animism on the island of Manus in New Guinea suggests that children do not readily commit to, or readily believe in, the counter-intuitive concepts they hear expressed by adults. In fact, they seem somewhat resistant to adopting such concepts. Using a wide range of methods, including both observational and experimental probes, Mead reports that while children up to age 12 can verbally repeat counterintuitive representations heard from adults (i.e., they remember them) children do not spontaneously use these concepts for themselves, do not readily deploy them in explaining novel objects with unexpected behaviors (e.g., dancing dolls, typewriters, Chinese wind chimes), and tend to ignore adult admonitions involving counter-intuitive agents or propositions. For example, adults believe that seeing one’s reflections in clear water should be avoided, as a water demon will steal part of one’s soul. Children aren’t taken on trips to the mainland because they won’t listen to direct admonitions about this, and end up staring into the fresh water. Children, however, will readily respond to admonitions about avoiding certain areas due to large crocodiles. In Manus, children would seem to possess a naturalistic bias, as predicted by the above theory. We will return to this work as it relates to religious representations in the General Discussion, where I will address what is one of the central challenges to this approach to religion: the Mickey Mouse Problem. The Mickey Mouse Problem is the idea that while many 3 While there is a rising tide of work showing that child and adults are resistant to beliefs that inviolate their intuitive expectations (Bloom and Weisberg, 2007), Mead’s findings may seem at odds with some other recent work in developmental psychology (e.g., Bering and Parker, 2006). To my knowledge, however, none of this work has been done with children in a socio-cultural environment like Manus, so any claims about human nature are necessarily proffered with caution. The industrialized environment where this work has been done is filled with talking animals (cartoons), flying men and machines (superheroes and planes), disembodied voices (phones), etc. Moreover, many of these counter-intuitive representations are specifically designed for and directed at children (unlike in places such as Manus). If there were a place where we’d expect children to accept counter-intuitive explanations, it would be this in bizarre (by the standards of human evolutionary history) socio-cultural environments of modern industrialized societies. Inferentially potent displays and cultural evolution 11 counterintuitive concepts, like God or Mickey Mouse, are highly culturally successful and inhabit the minds and memories of millions, there is an important difference: people, at least some people, are deeply committed to God, organize their lives around him/her/it, and might even die for their God. Mickey does not engender such commitments, despite Disney’s best efforts. So, the question is: How is God different from Mickey Mouse? Part I: The emergence of an adaptive challenge The evolution of high fidelity cultural learning, with all its adaptive benefits (Boyd and Richerson, 1995), increases the potential for exploitation by other members of one’s group because cultural learners are open to modifying their behavior (and underlying mental representations) in response to others’ behavior. Potential models can manipulate learners by displaying behaviors inconsistent with their true underlying mental representations. However, prior to the evolution of sophisticated forms of symbolic communication, of which language is the most relevant example, this potential was minimal since learners had to actually observe their model “in action” to acquire his or her practices, preferences, beliefs, or strategies. For example, in acquiring a particular tool making practice, learners had to watch their chosen models actually making the tools, and the final product testified, at least in part, to the effectiveness of the observed manufacturing practices. A model who wanted to deceive others about his favored technique could demonstrate a less effective technique in front of the learners, but this would be costly in time and effort, and the learner may not be fooled because in the end a less effective tool would result. Similarly, in acquiring food preferences (diet choice), pre-linguistic cultural learners presumably watched what foods others actually consumed, and how this food was found, extracted, and prepared. Manipulation in this case would require consuming a nonpreferred food, with all of its associated costs. Inferentially potent displays and cultural evolution 12 With the evolution of symbolic communication, in which mental representations (e.g., beliefs) can transmit at low cost, the opportunities for ‘Machiavellian manipulators’ to exploit learners would have dramatically increased. These manipulators hold one mental representation but express another (e.g., state it verbally) in an effort to cause others to do things that will increase the manipulators’ relative fitness. For example, a Machiavellian manipulator might believe that ‘blue mushrooms will make you sick in the long run’ and therefore he won’t eat them, but he verbally announces with enthusiasm that “blue mushrooms are quite tasty.” An unwitting cultural learner who has selected this prestigious Machiavellian as a model might then acquire the mental representation that ‘blue mushrooms are tasty’ and start eating them. Since we know from both theory and evidence that prestigious individuals can influence the beliefs (and other mental representations) of many learners, a prestigious Machiavellian could dramatically increase his relative fitness with well-designed culturally-transmitted mindviruses. With language at his disposal, this Machiavellian could manipulate existing beliefs. For example, people in many places believe “the wishes of our dead ancestors must be obeyed”. A manipulator might transmit the belief—not held by him—that he is “the mouthpiece for the ancestors, and they will talk through him; their first command is to pay the mouthpiece for his service to the ancestors with 1 pig from each house in the village.” Below, I show experimental evidence that learners are cognitively immunized against such persuasion. I hypothesize that natural selection addressed the emergent problem of Machiavellian manipulators, not by suppressing any use of symbolic communication in cultural learning, but by constructing a kind of cultural immune system. This immune system is designed to assess a potential model’s ‘degree of belief or commitment’ to a symbolically communicated belief using the model’s displays or actions. Cultural learners should look for displays that are most Inferentially potent displays and cultural evolution 13 consistent with the expressed representation(s) and, more importantly, look for actions that may indicate belief in something inconsistent with the model’s expressed beliefs. Such diagnostic actions are evidence of commitment to their expressed belief. A model, for example, might express the view that donating to charity is important, but not donate when given the opportunity. Such an action, failing to donate, should indicate to a learner that while the model may believe in some sense that giving to charity is a good idea, he’s probably not deeply committed to it. As we’ll see, cultural learners under such conditions would simply acquire the practice of talking about how good it is to give to charity, without actually practicing such giving. Learners imitate the model, in both actions (talking about how important charitable giving is) and in degree of commitment (not much). Diagnostic actions, such as actually giving at a cost to oneself, are inferentially potent displays. Often, these inferentially potent displays will appear costly if one holds one belief about the world, but substantially less costly, neutral, or even beneficial if one holds an alternative belief. In our mushroom example, the act of simply eating the blue mushroom would seem costly, and unlikely, if the model believes that blue mushrooms are poisonous but not costly at all if the model believes what he stated, that blue mushrooms are tasty. The action of actually eating the blue mushroom is an inferentially potent display for the verbal expression of the underlying mental representation of liking to eat blue mushrooms. The likelihood of eating the blue mushroom if one actually believes mushrooms are toxic is small. This approach does not mean that learners ignore verbal statements, or other forms of symbolic communication. Such symbolic expressions can be extremely informative in a learner’s efforts to replicate the underlying mental representations of a chosen model or models. Since context and content transmission biases don’t disappear in the absence of inferentially potent Inferentially potent displays and cultural evolution 14 displays, cultural learners will still recall the verbal statements of, for example, prestigious individuals better than the statements of others. The key is that, in the absence of inferentially potent displays, learners are not committed to those recalled representations in a manner that propels behavior beyond simply repeating the verbal expression itself. Psychological findings The evolutionary logic laid out above proposes that learners ought be more likely to acquire a culturally-transmitted mental representation, in the form of practices, beliefs, values, strategies, etc. if their potential models perform acts that are both consistent with the possession of the underlying mental representation (which might be expressed verbally) and not consistent with holding an alternative mental representation. Stated another way: If identical cultural models verbally express the same belief, preference or opinion, learners are—ceteris paribus— more likely to learn from models who perform accompanying inferentially potent displays. Often, the more costly a model’s display would seem to someone who did not hold the expressed belief, the greater the influence of that model on learners’ subsequent commitment to, or belief in, the expressed representation. Here I unite findings from three seemingly independent areas of psychology, all of which study cultural learning in one form or another. These research programs focus on the cultural transmission of (1) food preferences, (2) opinions, and (3) altruism and social behavior. Acquisition of beliefs, attitudes and behaviours in all three of these areas has already been shown to be influenced by prestige-biased transmission and/or conformist transmission. The question addressed here is whether learning in these areas also show evidence of influence from inferentially potent displays. Inferentially potent displays and cultural evolution 15 Food preference People’s preferences for certain foods, as well as the amount of food they consume, is substantially influenced by which foods those around them prefer and how much they eat. In research with children, findings indicate that learners actually shift their intrinsic food preferences toward those of their models, especially when those model are older and of the same sex (Birch, 1980, 1987; Duncker, 1938). Work with adults demonstrates that models can influence the quantity consumed, and that this may be the most important factor in quantity eaten (Herman et al., n.d.; Herman et al., 2003). These data indicate that food is governed by the kinds of adaptive cultural learning biases that govern other domains of culture. If food is also governed by inferentially potent displays, then learners should be more inclined to eat novel foods when a model is first observed to eat the food himself. As in our blue mushroom example, eating is an inferentially potent display for believing something is worthy of eating (or at least not toxic). Harper and Sanders (1975) report experimental finding in which a female experimenter went to the homes of children (ages 14 to 48 months), spent at least 20 minutes playing with the child until he or she seemed comfortable, and then presented the children with a novel food. In the baseline treatment, the experimenter merely placed the novel food out (within reach of the child) and said “something to eat” to the child. In the inferentially potent display treatment, the experimenter said the same thing as she sampled some of the food. In the baseline only 25% of children tasted the food while in the second treatment, 75% sampled (p < 0.05). This may seem both intuitive and unsurprising (it should), but I will argue it 4 The findings also reveal the expected sex-bias, with girls copying the female experimenter more than boys. Also note that while non-human primates do show some effects of social facilitation on trying novel foods (they are more likely to try a novel food if others are eating something), human children focus on what is actually being consumed by others, and will only sample the novel food if the others are eating the same thing (Addessi et al., 2005; Visalberghi and Addessi, 2001; Visalberghi et al., 2003). Inferentially potent displays and cultural evolution 16 represents a manifestation of a tendency for cultural learners to look for displays in potential models that indicate the model actually believes what he or she is saying. Opinion transmission Under the rubric of persuasion, psychologists have long studied both the characteristics of effective “communicators” and the contexts of opinion change (Tannenbaum, 1956). From the evolutionary perspective presented in this paper, persuasion or opinion change is merely one kind of cultural transmission. When models express something verbally (or in writing), ostensibly their own underlying mental representations, this may cause others to alter their own mental representations in an effort to move closer to the representation that they perceived from a model. As predicted by the above approach, opinion change research shows that subjects show greater opinion shift both when the model is more prestigious and when the frequency of others expressing similar opinions is greater (conformist transmission; people move their opinion toward that of the majority). This same work also shows evidence of inferentially potent displays, although in a more nuanced manner than with food. Walster el. al. (1966) had subjects read newspaper articles in which either a high prestige individual (a famed prosecutor) or a low prestige individual (a thug) expressed opinions about the need for changes in the criminal justice system. Their opinions called for changes that would run either for or against their own self interest. Opinion measures from the subjects show that when models’ expressed opinions that promoted their own interests, subjects’ opinions shifted toward the model substantially less than when models expressed an opinion contrary to their own (the models’) interests. Here, the inferentially potent display is the verbal opinion itself. It’s inferentially potent in this context because the dissemination of the expressed opinion, which was given to the mass media (newspaper article), runs against the self-interest of the model. It Inferentially potent displays and cultural evolution 17 seems unlikely that a model would express an opinion counter to his self-interest if he actually held an opinion consistent with his self-interest. The evidence also suggests that the influence of high-prestige individuals is damaged more when they advocate for their own interests than when low-prestige individuals advocate for their own interests. In fact, when a low prestige individual advocates for a view that runs counter to his self-interest, his influence exceeds that of a high prestige individual advocating for a view favoring his self interest. Eagley et. al. (1978) provides similar findings. As mentioned earlier, these finding demonstrate that our adaptation for using inferentially potent displays has been calibrated to recognize that high prestige individuals have more incentive to make self-serving claims, since their opinions are more likely to spread. Cultural transmission of altruism requires acts of altruism Research with children on the social learning of altruism toward anonymous others shows that a model’s verbal statements (“exhortations” or “preaching”) to make charitable donations have little or no impact on learners unless such statement are accompanied by the model actually making costly donations himself. Once the model donates, cultural learning powerfully transmits altruistic behavior or charitable preferences. Actually donating is an inferentially potent display that would be unlikely to be observed if the model held beliefs or preferences about charitable giving substantially different from those he expressed verbally. In the paradigmatic experimental setup, from which there have been numerous variations, a child is brought alone to the experimental area (often a trailer on school grounds) to get acquainted with the experimenter. Then, the child is introduced to a bowling game and shown a range of attractive prizes that he or she can obtain with the tokens (or pennies or gift certificates) won during the bowling game. The subject is also shown the charity jar for “poor children” Inferentially potent displays and cultural evolution 18 where they can put some of their winnings, if they want. This jar is next to the bowling game and often has a “March of Dimes” poster over it, or some facsimile. A ‘model’, who could be a young adult or another peer (though usually not the experimenter), demonstrates the game by playing 10 or 20 rounds. On winning rounds, which are pre-set, the model donates (or not, depending on the treatment) to the charity jar. After finishing the demonstration, the model departs (or not, depending on the treatment) and the child is left to play the bowling game alone, often monitored through a one-way mirror. As background, the results from numerous researchers involving hundreds of children (ages 5 to 11) and a wide range of experimental variations, demonstrate three robust findings. First, children spontaneously acquire either the generosity or selfishness of the model. Compared to the amount that children donate in the absence of a model, children donate more to the charity jar if they see a model donate and they donate less if they saw a model fail to donate. The more the model donates the more the children donate, and the more opportunities a child has to observe the model the greater the degree of transmission (Bryan, 1971; Bryan and Walbek, 1970b; Grusec, 1971; Presbie and Coiteux, 1971). Presence of the model during the donation phase of the experiment has little effect (Rosenhan and White, 1967). Third, these effects endure over months in retests (without a model present) and extend to similar contexts (Elliot and Vasta, 1970; Midlarsky and Bryan, 1972; Rice and Grusec, 1975; Rushton, 1975). Specific to our interests here, several studies compare the effect and interaction of models who preach generosity or selfish (“one ought to donate...”) and practice either generous or selfish giving. Preaching alone usually has little or no effect on giving. Children’s behavior seems uninfluenced by preaching when these exhortations are inconsistent with the model’s actions (Bryan et al., 1971; Bryan and Walbek, 1970a; Bryan and Walbek, 1970b; Rice and Inferentially potent displays and cultural evolution 19 Grusec, 1975; Rushton, 1975). When a model actually donates generously, the subjects donate more generously. Here, giving away tokens that one could use to exchange for toys is an inferentially potent display of one’s commitment to the verbal claim that “one ought to donate.” Researchers have looked for, but not found, a detrimental effect on acquiring altruism from hypocritical models. Researchers hypothesized that children may give less than they would in a no-model control after observing a model that said one thing (e.g., “give a lot”) and did another (gave nothing). Instead, the children just seem to give little weight to a model’s words, in the absence of an inferentially potent display (Bryan et al., 1971; Midlarsky et al., 1973). Verbal expressions, however, if they are accompanied by inferentially potent displays help the learner figure out the underlying details of the model’s mental representations—that is, the where, who and why of charitable giving. Experimental work shows that exhortations combined with inferentially potent displays allow learners to broaden the range of contexts for acquired altruism (Grusec et al., 1978). Thus, verbal expressions can be critical to understanding what is learned, but these young learners seem to “switch off” unless such verbal statement about what one ought to do, when, and why, were accompanied by an inferentially potent display. Such empirical findings, which applied to both children and adults, suggest that our capacities for cultural learning may have been shaped to weigh a model’s inferentially potent displays in adopting and committing to culturally transmitted representations. Substantially more research is needed to understand precisely how this aspect of cognition works. Now I turn to the implications of this cognitive learning bias for cultural evolution. Part II: Do inferentially potent displays affect cultural evolution? If indeed our species is endowed with a cognitive learning mechanism that weighs the presence of inferentially potent displays in cultural learning, what implications does this have for Inferentially potent displays and cultural evolution 20 cultural evolution? How might this influence the kinds of stable cultural phenomena we observe across societies? Could it explain the widespread and unusual nature of the kinds of costly displays we highlighted at the outset, such as animal sacrifice, subincision, scarification, selfmutilation, or tattooing? Here, I begin to explore this possibility by constructing a simple cultural evolutionary model that incorporates the effects of inferentially potent displays. Analytical model Building on typical cultural evolutionary modeling approaches (McElreath and Boyd, forthcoming), this model adds a cognitive mechanism that weighs inferentially potent displays to the standard use of success-biased transmission. Cultural learners, in figuring out who to learn from, consider both a potential model’s success and whether their expressed belief is also supported by an inferentially potent display. The best evolutionary models strip away as many details as possible in order to focus on the internal governing dynamics. To this end, my model focuses on the cultural coevolution of two different kinds of mental representations, a belief (θ) and practice or display (x). For simplicity, the model assumes that both θ and x are discrete dichotomous variables, taking on values of either 0 or 1. What I mean by belief and practice/display will become clear as we proceed. In the principle situation under investigation, the two belief variants of θ (0 or 1) have identical content and context biases. This means both that (1) possessing a particular value of θ does not directly affect an individual’s success or their attractiveness as a model, and (2) no aspects of the representational content of variants 0 or 1 make any relative differences in their This simple model submerges the difference between knowing a belief is out there (e.g. that others believe it) and committing deeply to a belief. If an individual has θ = 1, he believes in and is deeply committed to whatever this belief represents. If an individual has θ = 0, he is not committed to this belief, but he may or may not know that others are committed to such a belief. Inferentially potent displays and cultural evolution 21 likelihood of transmission. In terms of its direct effects, θ is neutral. The belief θ is a mental representation (e.g., God is watching) that is expressed verbally, and without cost. In contrast, the mental representation x is a practice that does influence success: individuals with x = 1 can be thought of as performing a costly act (e.g., attending long boring Sunday rituals or getting a painful tattoo) while those with representation x = 0 pay no costs (e.g., not attending Sunday rituals, or not getting a tattoo). However, the variants x = 1 and θ = 1 are linked in two interrelated cognitive senses. First, x = 1 is an inferentially potent display for θ = 1, meaning that if someone displays x = 1 and expresses θ =1, a learner will be more likely to acquire θ = 1 than he would if his model had displayed x = 0. A cultural learner observing a prestigious model who consistently attends those boring rituals and says “God is watching” is—ceteris paribus—more likely to acquire the idea that “God is watching” (or code “'God is watching' is true,” see Bergstrom et al., 2006). Second, individuals possessing θ = 1 have a content (e.g., compatibility) bias for acquiring variant x = 1. This means that if you believe that “God is watching” (θ = 1) you are more susceptible to acquiring the practice of attending Sunday rituals (x = 1) than if you hold the belief θ = 0 (“God is not watching”). While here I used a content bias to model the link between having θ = 1 and acquiring x = 1, there are other ways to think about how having θ = 1 could influence performing x = 1. These are described at the end of this section. To illustrate all this, consider this simplified example. Suppose people with θ = 1 deeply believe in, and are committed to, the idea that eating mostly high protein vegetable foods will improve long-term health, fitness, and attractiveness. Those with θ = 0 don’t believe this (although some do profit from selling tofu). Further, suppose that those with x = 1 eat lots of very expensive, high protein tofu, and those with x = 0 do not. When our adaptive cultural learner meets a prestigious model who is observed only to verbally express his belief (θ = 1) in Inferentially potent displays and cultural evolution 22 the value of eating high protein vegetable foods he substantially devalues this model in deciding whether to change his θ belief to 1. However, if our learner also sees this prestigious model eating tofu (x = 1), he does not devalue the model as much in deciding to acquire the model’s belief. Observing a potential model eating a lot of tofu (x = 1) is inferentially potent for a belief that vegetable protein is important for health, etc. because—let’s assume—(1) few people would actually eat tofu (x =1) without some supporting belief in its health consequences (θ = 1) and (2) eating tofu is perfectly consistent with believing θ = 1. Note that all models are “devalued” (weighted less) relative to oneself due to the uncertainty of others’ true mental representations vis-à-vis one’s own. With regard to acquiring x (deciding what to eat), individuals who believe (θ = 1) that eating high protein vegetable foods is the key long-term health, fitness and attractiveness will find the practice of eating lots of tofu (x =1) more attractive than those who believe θ = 0 (who experience only the costs of the bland mushy taste). To formalize this, I sought to minimally modify the standard approach to cultural evolutionary modeling, using replicator dynamics, in order to build incrementally on a well understood approach. The transmission of both beliefs (θ) and practices (x) assumes that during each time step a learner encounters one potential model. If the model expresses variants that are the same as those already possessed by the learner, the learner does not modify his cultural mental representations. However, if the learner and model differ, the learner changes his variants with a probability related to the difference in the learner’s own weighting and that of the model. For the transmission of θ, the weighting of the model will be influenced by both her success and by the presence or absence of the inferentially potent display (x = 1). To keep things simple, this 6 In general, the differences in weightings arise from the learner’s perceived differences between himself and the model in success, sex, age, skill, etc. and any content biases. Inferentially potent displays and cultural evolution 23 analysis assumes that the underlying belief in question is neutral with regard to success (as mentioned above), so only the value of x influences the transmission of θ. Models with x =1 have a weighting of ω c, where c is the cost of the practice x = 1 and ω is the base weight assigned to all individuals. Models with x = 0 have only ω for their weight. Since weights must be greater than zero, ω > c. The learner adjusts the weighting of the model depending on the model’s observed practices (x). If the model holds the belief-practice combination of 1/0 or 0/1 the weight of the model is adjusted by a factor of (1σ), where σ is between zero and 1. If the model holds a belief/practice combination of 1/1 the weight of the model is adjusted by a factor of (1σ + ψ). If the model possesses a belief/practice combination of 0/0 the weight of the model is adjusted by a factor of (1-σ + δ). The parameter σ captures a generalized skepticism towards acquiring beliefs that are cheaply expressed symbolically, while ψ and δ capture the extra inferential evidence provided by the presence of x = 1 for acquiring θ =1 and for x=0 for acquiring θ =0. In our toy tofu example above, a model who expresses the belief that eating high protein vegetable food is highly beneficial and is observed actually eating tofu (x = 1) suffers less de-weighting than models with other belief/practice combinations—ψ ≥ δ ≥ 0. Since the adjustment of the model’s weighting is meant to capture the learner’s uncertainty about the model’s actual underlying belief (θ), no adjustment is applied to the learner’s own weighting (the learner knows, in some sense, his own beliefs: σ = δ = ψ = 0 for learner’s weighting). For the transmission of x, all individuals with x = 1 will experience the same cost, c, as above, but those learners with belief θ = 1 will also experience an attractiveness, b, for the content of the practice x = 1, giving models holding the belief/practice combination 1/1 a weight Inferentially potent displays and cultural evolution 24 of ω -c + b. Since practices or displays are not symbolically (and cheaply) displayed, no inferentially adjustments need be applied to the success weights (σ = δ = ψ = 0). Using the above assumptions, along with φ to track the frequency of individuals with belief θ = 1 and q for the frequency of individuals with x = 1 in the population, two expressions emerge, one for the change in φ (Δφ) and another for the change in q (Δq). [ ] c b q q q − − = Δ φ β ) 1 ( (1) [ ] q c q q ψ δ ψ ω φ βφ φ − − − − = Δ )) 1 ( ( ) 1 ( 2 1 (2) β in each of the above equations is a positive constant that expresses how learners convert weighting into the probabilities of changing cultural variants and guarantees that difference in the weights multiplied by β does not exceed 1. The larger β is, the more individuals weight any particular learning encounter. The terms q(1-q) and φ(1-φ) are also always nonnegative and express the variance in φ and q, respectively. For simplicity, we’ll set ω = 1. A complete analysis of this system shows that there are three potential stable situations, one in which only the no-cost (φ = q = 0) equilibrium is stable, one in which only the costly (φ = q = 1) equilibrium is stable, and a third in which both of these equilibria are stable at the same time. The first (φ = q = 0) equilibrium is the sole equilibrium if either of the following conditions are met: b ≤ c or ψ = δ =0. This essentially replicates existing work: without inferentially potent displays costly practices don’t have a stable equilibrium. Note that, even if b > c, the costly practice still cannot be sustained unless practices are inferentially potent. In the second situation only the costly equilibrium (φ = q = 1) is stable. This can occur if ψ > δ = 0 and b > c. For this stable equilibrium to exist, the x = 0 display must provide the learner with no hint that the model is more likely to believe θ = 0 than θ = 1. Given that this Inferentially potent displays and cultural evolution 25 equilibrium also requires that b > c, which tends to link x = 1 and θ = 1, such an equilibrium might only exist under very specialized conditions. The third situation involves two stable equilibria, the no-cost one and the costly one, as illustrated in the phase plot in Figure 2. This occurs when (3) and (4) are both satisfied (this assumes ψ and δ > 0).