Original Paper The evolution of social play by learning to cooperate Adaptive Behavior 1–14 Ó The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/1059712315608243 adb.sagepub.com Sabine Durand1 and Jeffrey C Schank1,2 Abstract Social play is common in mammals but its adaptive significance is not well understood. A commonly held hypothesis is that social play allows juveniles to learn skills and rules for cooperation as adults. On this view, the adaptive benefit of social play derives from the benefits of cooperation as adults. However, cooperation is only beneficial if it is used in populations of predominantly cooperators; otherwise, it is a costly strategy. We investigated the latter problem by modeling the link between social play and subsequent adult cooperation using an agent-based model. In our model, agents had a play gene with allelic social play states (i.e., play or not play). Juveniles with a play gene learned to cooperate by successfully engaging in social play with another juvenile agent with a play gene. As adults, agents played the stag hunt game with other adults to obtain resources for reproduction. Those that learned to cooperate by playing as juveniles cooperated in the stag hunt game. When agents aggregated into small groups, we found that social play could evolve by facilitating the learning of cooperation. Our theoretical results also show that social play is a novel mechanism for the indirect evolution of cooperation. Keywords Evolution of cooperation, social play, learning to cooperate, evolutionary simulations, fairness, population structure 1. Introduction Social play is common in mammals (e.g., Bekoff, 1974; Gomendio, 1988; Hayaki, 1985; Mackey, Makecha, & Kuczaj, 2014; Thor & Holloway, 1985; Vidya & Sukumar, 2005; Watson, 1998) and comes in many forms such as chasing, wrestling and play-fighting, and in some species more complex forms of social play such as sociodramatic play in human children (Graham & Burghardt, 2010). From an evolutionary perspective, social play has potential costs such as increased exposure of young animals engaged in social play to predators, injuries, and energy costs (Graham & Burghardt, 2010). There are several hypotheses regarding the adaptive benefits of social play, including play as practice for adult behaviors (e.g., fighting or mating) and play as a way to develop motor skills (Graham & Burghardt, 2010). Evidence both for and against these hypotheses has been found but there are no definitive explanations for the adaptive benefits of play (Caro, 1980; Nunes et al., 2004; Sharpe, 2005). It is likely that play serves many functions. In this model, we focus on one functional hypothesis: social play as a means by which juveniles learn the skills and rules needed for cooperation in adulthood. Social play theorists have hypothesized that social play gains delayed adaptive benefit by facilitating the learning of adult social skills such as cooperation by practicing the skills required to socially interact and cooperate as adults (Bekoff, 2001; Bekoff & Pierce, 2009; Fagen, 1981; Lee, 1983; Pellis & Pellis, 2007; Sussman, Garber, & Cheverud, 2005). During social play, individuals may learn rules of conduct and gain an understanding of what is and is not acceptable to others—skills that may later be applied in cooperative contexts (Dugatkin & Bekoff, 2003). Indeed, cooperative play has evolved in a number of species (Bekoff, 1995; Bekoff & Allen, 1998; Fagen, 1981; Power, 2000). For example, animals may learn how roughly they can interact with others and how to solve conflicts during play (Dugatkin & Bekoff, 2003). For social play to be adaptive, the delayed benefits of learning to play must outweigh the costs of social play. However, cooperating in a population of non-cooperators is costly. If the 1 Department of Psychology, University of California, Davis, California, USA 2 Animal Behavior Graduate Group, University of California, Davis, California, USA Corresponding author: Jeffrey C Schank, Department of Psychology, University of California, One Shields Ave., Davis, CA 47405, USA Email: [email protected] Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 2 Adaptive Behavior benefits of adult cooperation are to explain the evolution of social play, then social play must evolve in a way that facilitates the success of adult cooperation. Here, we focus on how an evolutionary positive feedback loop can emerge in which the evolution of social play facilitates the success of adult cooperation and the success of adult cooperation feeds back to facilitate the evolution of social play. The potential fitness costs of social play are clear. For example, Harcourt (1991) found that South American fur seal pups only spend about 6% of their time playing, yet 85% of seal pups that were taken by predatory sea lions were playing at the time they were captured. When engaged in play behavior, marmots have been found to take longer to respond to predators, resulting in greater predation risk (Blumstein, 1998). Injuries are not uncommon during play. Bighorn sheep incur injuries by impaling themselves on cactus spines and elephants often get trapped in mud during play (Berger, 1980; Douglas-Hamilton & DouglasHamilton, 1975). With regard to energy costs, Miller and Byers (1991) have found that playing in fawns consumes approximately 20% of their daily energy expenditure in excess of resting metabolic rate and growth. Domestic kittens were estimated to spend approximately 4–9% of their total daily energy expenditure, excluding growth, on play (Martin, 1984). Furthermore, play behavior can impede the acquisition of food. Caro (1995) found that playing in cheetah cubs significantly decreased maternal hunting success, as the cubs’ play alerts prey of their mother’s presence. If social play is beneficial, its benefits must outweigh costs due to exposure of young animals to predators, injuries, and energy expenditure. As the costs of social play are suffered early in development, the benefits delayed until adulthood must be substantial. But even if the benefits of adult cooperation are substantial, they are only reaped if other adults successfully cooperate. If an individual attempts to cooperate in a population of non-cooperators, there is a net cost to cooperation rather than a benefit (Axelrod & Hamilton, 1981). Thus, contrary to the adaptive benefits of cooperation hypothesis for the evolution of social play, it appears that cooperation is unlikely to be beneficial unless a population already has evolved cooperation. There is some evidence suggesting that juvenile animals learn to cooperate and behave fairly through social play (Bekoff, 2001). During social play, juveniles must learn social behaviors and rules that are acceptable to others, or risk being excluded by other juveniles from play (Bekoff, 1974). In canids, pups avoid playing with individuals that are too rough (Brosnan & de Waal, 2012). As a result, juveniles often engage in fair play behaviors during social play (Dugatkin & Bekoff, 2003). Individuals will sometimes exhibit selfhandicapping behavior, in which they voluntarily put themselves in vulnerable positions, giving their play partners the advantage (Bauer & Smuts, 2007). For example, red-necked wallabies modify their play behavior based on their playmate’s age, using gentler play behavior with their younger, less experienced, playmates (Watson & Croft, 1996). Many rodent and primate species have been observed to forgo the inclusion of a defensive maneuver in their attacks when playfighting, unlike in serious fight scenarios, allowing their play partner an opportunity to counterattack (Pellis, Pellis, & Reinhart, 2010). Play partners that are on the receiving end of the attack may also be slower on the defense during play fights, allowing their play partner to take the advantage (Pellis et al., 2010). The rules of conduct and understanding of fairness and cooperation learned during social play may carry over to future social interactions, such as cooperative interactions, in adulthood (Bekoff, 2001). Several studies have provided evidence suggesting social play experiences in early development are important for establishing competent social skills in adulthood (Einon et al., 1978; Hol, Van den Berg, Van Ree, & Spruijt, 1999; Pellegrini, 1992; Pellis & Pellis, 2007). Rat pups deprived of social play were found to display significantly reduced social behavior as adults (Hol et al., 1999; Van den Berg et al., 1999). Social play deprivation in rats is also associated with difficulty coordinating movements with a partner in adult social interactions and a tendency to exhibit hyperdefensiveness in response to an approaching rat (Pellis & Pellis, 2007; Pellis, Field, & Whishaw, 1999). However, socially isolated rats that were permitted short daily play sessions exhibited relatively normal social behaviors in adulthood (Einon et al., 1978). Though little research has been done in humans, Pellegrini (1992) found that children who engaged in more rough and tumble play tended to exhibit better social problem solving abilities. Social play has been found to be relevant in the development of a variety of social skills—including initiating social interactions, coordination with a partner, and the ability to act appropriately around peers—that are likely necessary for adult cooperation (Pellegrini, Dupuis, & Smith, 2007). Yet, there is also evidence against the hypothesis of social play as learning for adult cooperation. For example, grizzly bears and orangutans engage in social play extensively during their juvenile stage, but are solitary animals that typically do not cooperate with one another in adulthood (Burghardt, 1982). Baldwin and Baldwin (1974) found that squirrel monkeys can develop normal social organization and social behavior in the absence of social play, though individuals that engaged in more social play were found to have improved social skills. Empirical results suggest that juveniles can learn to cooperate and behave fairly by engaging in social play and that what they learn can facilitate adult cooperative Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 3 behavior. There are, however, theoretical problems with the cooperation hypothesis. If social play is introduced in a population of non-cooperators at low frequency and social play derives its benefits from social cooperation, it would appear theoretically impossible for social play to evolve. Moreover, if social play is introduced at low frequency into a population, how do juveniles find other juveniles that will engage them in social play so that they learn to cooperate as adults? The theoretical problem of explaining social play via adaptive benefits of adult cooperation appears to be a vicious circle: there must be sufficiently many juveniles with heritable disposition for social play, but if cooperation as adults is not favored at low frequency, the number of individuals with a disposition towards social play cannot increase. To break this circle, we show that population structure together with social play can result in small spatial clusters of agents with high frequencies of juveniles that can engage in social play and cooperate as adults. To investigate this possible explanation, we developed an agentbased model of the evolution of social play that links social play to cooperation via social learning during development. 2. Model Our agent-based modeling approach emphasizes modeling generic properties of organisms relevant to the theoretical problem of interest (Schank, Smaldino, & Miller, 2015), which more broadly can be characterized as synthetic ethology (MacLennan & Burghardt, 1993). From this perspective, we begin by supposing that agents go through two developmental stages: a juvenile stage and an adult stage. During the juvenile stage, juveniles that have a genetic disposition for social play can engage in play with one another and learn to cooperate via bouts of mutual social play. As adults, agents with the necessary social play experience cooperate with other adults in obtaining resources. Suppose that (1) at both juvenile and adult stages, agents attempt to aggregate with others of the same developmental stage, (2) when adult agents achieve sufficient resources, they can reproduce, and (3) dispersal of their offspring is limited. These conditions will produce structured populations of groups of agents. Because it takes at least two agents to engage in social play as juveniles and thereby learn to cooperate, cooperators will also likely be in spatiotemporal proximity with each other as adults. In such a scenario, small groups of two or more cooperators may occasionally appear (via learning to cooperate by social play) and spread throughout the population. 2.1 Adult cooperation Consider a population of adult agents that play the stag hunt game (SH) for resources. The SH is based on a scenario first described by Jean-Jacques Rousseau, in which two individuals go hunting (Rousseau, 1992). They can each hunt a small hare on their own for a small meal, but if they would like a stag for a larger meal, the two individuals must join forces. If both players cooperate and hunt stag, both gain the highest payoff possible, but if one defects and hunts a hare instead, the other is left with nothing for its efforts. The SH is a type coordination game in which agents must coordinate their behavior to achieve a cooperative gain and is most appropriate for the initial evolution of cooperation in a population (Skyrms, 2004). More formally, we assume non-cooperators each receive a benefit b each time they play minus a cost c. By cooperating, their joint gain is increased by a factor x. If two cooperators (C) play, they split their cooperative gains, xb, and each receive xb/22c units of resource. If two non-cooperators (N) play, they do not cooperate and each gains b2c, which is the quantity of resources they can gain on their own. If a cooperator and non-cooperator play, the non-cooperator gains b2c because it does not cooperate, but the cooperator receives only –c, which is no resource minus the cost. When x.2, two evolutionarily stable strategies (ESS) exist for the SH. This implies that there exists a fixed point, p, below which the frequency of cooperators evolves to p=0 and above which the frequency of cooperators evolves to p=1. To find this fixed point we let p represent the frequency of cooperators in a population and then write a differential equation specifying how p evolves over time dp = p(E(C) E(N)) dt ð1Þ where E(C) is the expected payoff for cooperators and E(N) is the expected payoff for non-cooperators. The expected payoffs are given by Equations 2 and 3. xb c ð1 pÞc E ðC Þ = p 2 ð2Þ EðN Þ = pðb cÞ + ð1 pÞðb cÞ = b c ð3Þ By substituting Equations 2 and 3 into Equation 1 we obtain dp xb = pð1 pÞ p b dt 2 Solving for ð4Þ dp = 0 yields the fixed points p=0, p=1, dt 2 if x.2. x For our simulations, we set b=1, c=0, and x=4 with an unstable fixed point of p=1/2. If the proportion of cooperators, p, is less than 0.5 (i.e., in the ESS zone for non-cooperation), the frequency of cooperators will decrease to the stable fixed point p=0. If the proportion of cooperators, p, is greater than 0.5 (i.e., in and p = Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 4 Adaptive Behavior the ESS zone for cooperation), the frequency of cooperators will increase to the stable fixed point p=1. If agents can play any other agent in the population with equal probability, cooperators cannot invade a population of non-cooperators when introduced at low frequencies. However, consider a population of agents that on each round of play randomly play another adult agent from an adjacent cell rather than an agent from the population as a whole. Agents in this population play other agents in their spatial proximity. Thus, even when cooperators are introduced at low frequencies in a population, random drift and mutation can result in spatial regions that have higher frequencies of cooperators than others. In such regions, cooperators do relatively better. 2.2 Space and movement learn to cooperate. The more they engage in social play, the more likely it is that they cooperate as adults. The adult stage lasts Li2D. 2.4 Learning to cooperate When juvenile agents mature into adults, their probability of cooperating in the SH depends on the cooperation they have learned by repeatedly engaging in social play (Figure 1). Learning to cooperate by engaging in social play was modeled using a version of RescorlaWagner (1972) learning model given by Equations 5 VCP ðt = 0Þ = 0 VCP ðtÞ + a½1 VCP ðtÞ if social play VCP ðt + 1Þ = if no social play VCP ðtÞ ð5Þ Agents are located in an X 3 Y square grid of cells with periodic boundaries. Each agent occupies only one cell; no two agents can occupy the same cell. On each round of play, agents search nearby cells (with radius SR around the agent) for other agents of the same developmental stage (adult or juvenile). An agent remains in its current cell if it is next to another of the same developmental stage. If an agent is not next to another agent, it will move to a randomly selected empty cell next to another agent of the same developmental stage that is within it search radius, SR. If there are no other agents at the same developmental stage within its search radius, the agent will move one cell using a zigzag random walk (Smaldino & Schank, 2012). 2.3 Agents An agent has a lifespan, Li, consisting of social play followed by adult play with the SH. The lifespan of each agent is determined by generating a number from a random normal distribution with mean L, standard deviation SDL, and rounding it to the nearest integer. The standard deviation, SDL, was selected to approximate a survival curve from the Gompertz standard model commonly used to represent age-dependent mortality rates in a wide variety of species (Easton & Hirsch, 2008). When adult agents reach the end of their lifespan, they die, as do their juvenile offspring, as we assume that juvenile development depends on parental resources. Adult agents may die earlier if they run out of resources. During the lifespan of an agent, the juvenile stage lasts exactly D=50 rounds. During the juvenile stage, agents receive a proportion, pc, of their parent i’s resources, Rit on each round of play. They also have the opportunity to engage in social play with other juvenile agents, but they can only engage in social play if both have at least one allele for social play. If two juvenile agents successfully engage in social play, they where a is learning rate and VCP(t) is the strength of association (1<VCP(t)0) between cooperation as an adult and bouts of social play. VCP(t) can only increase when two agents engage in social play and the probability, pc, of an agent cooperating as an adult is the value of VCP(t=D) at the end of the juvenile developmental period. 2.5 Reproduction and social play At any time during the adult stage, if an adult agent acquires sufficient resources to reach or surpass the reproductive threshold, RT, it attempts to reproduce a juvenile agent in an empty cell next to it. Agents can only reproduce if the current population size is less than the maximum population size, Nmax. If an agent successfully reproduces, it provides its offspring with a proportion, pI, of its resources (i.e., initial parental investment), and then subsequently provides a percentage, pc, of its resources to all of its juvenile offspring during each round of play until each offspring matures to adulthood. We investigate both asexual and sexual reproduction. For asexually reproducing agents, offspring inherit a social play gene with two allelic states (i.e., play or not play). If the social play gene is in the play state, then a juvenile agent attempts to engage in social play with other juvenile agents that it is next to. If the social play gene is in the not play state, then it does not attempt to engage in social play. At the time of reproduction, mutations occur at rate r and a mutation flips the social play gene into the opposite allelic social play state of its parent. Sexually reproducing agents are hermaphroditic and an agent can assume either a female or male role (but not both) during each pairwise reproductive episode. Hermaphroditic agents are used to investigate the role of sexual reproduction on the evolution of social play Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 5 and cooperation while at the same time maintaining populations with effective population sizes similar to asexually reproducing agents. Upon acquiring sufficient resources to reproduce, a randomly selected adult agent assumes a female role and searches for an available mate (another adult agent nearby that has also satisfied the minimum resource requirement for reproduction) to assume the male role. If a mate is successfully found, each agent contributes one allele. Each allele can mutate at rate r, which flips the allele into the opposite social play state of its parent. Thus, when a juvenile offspring is produced, it inherits one social play allele from each parent; each allele has one of two states (i.e., play or not play). For sexual reproduction, we modeled three types of allelic interactions: dominant, recessive, and additive. For the dominant case, if a juvenile has at least one social play allele, it will attempt to engage in social play with other juveniles and can learn to cooperate. In the recessive case, a juvenile must have both social play genes to engage in social play and learn to cooperate. In the additive case, juvenile agents with one social play allele engage in social play with probability 0.5, whereas juvenile agents with both social play alleles engage in social play with probability 1.0. For sexually reproducing agents, parental investment can be either uniparental (i.e., only the agent assuming the female role contributes resources to the offspring) or biparental (i.e., both parents contribute resources to the offspring). For the uniparental case, the agent assuming the female role provides its offspring with a proportion, pI, of its resources (i.e., initial parental investment), and then provides a percentage, pc, of its resources to the juvenile offspring on each time step. In the biparental care case, both reproducing agents provide juvenile offspring with a proportion, pI/2, of their resources followed by a percentage, pc/2, of their resources to their juvenile offspring on each time step. 2.6 Events On each time step of an agent’s life span, it is either a juvenile or an adult (Figure 2). If it is a juvenile, it dies if it has no parent(s) to provide resources. If it is old enough, it matures into an adult. If it has the play allele, it next searches for another juvenile agent with which to play; otherwise, it does nothing. If it searches and finds another agent that has not played, it attempts to engage in social play. If the other agent has the social play allele, they both engage in social play; otherwise, they do not. If it engaged in social play, it incrementally learns to cooperate as an adult. If an agent is an adult, it dies if it has no resources. Otherwise, it attempts to engage another agent next to it that has not yet played the SH that time step. The likelihood that both agents cooperate is a product of how much cooperation each agent learned as a juvenile. Figure 1. An illustration of juvenile learning rates during social play. During the juvenile period, agents with play genes engage in social play and depending on their learning rate, α, they increase their probability of cooperating as adults as a function of the number of bouts of social play. Whether or not an agent plays the SH, it can reproduce if it has sufficient resources. If it has sufficient resources, the population size is not at maximum Nmax, and there is an empty cell next to it, it can reproduce either asexually or sexually. 3. Evolutionary simulations The simulation model was written in Java using MASON, a multi-agent simulation package (Luke et al., 2005). Mason’s simulation engine supports asynchronous updating of agents, where the order of agent updating is randomized for each time step (see Table 1 for all fixed parameter values). 3.1 Initial conditions All simulations began with a population of adult agents that was half the maximum population size. Agents were placed randomly on a 50 3 50 2D space with periodic boundaries. The starting age of each agent was determined as described above. The starting resources of each agent were determined by generating a number from a random normal distribution with mean SR, standard deviation SDSR, and rounding to the nearest integer (see Table 3 for initial condition values). 3.2 Experimental conditions We systematically simulated two reproductive conditions: sexual and asexual reproduction. Within the sexual reproduction conditions, we systematically varied the rate of learning, maximum population size (or density, holding grid size constant), mode of inheritance (dominant, additive, and recessive), and type of parental care (biparental or uniparental) for a total of 5 3 3 3 3 3 2=90 conditions (Table 2). For the asexual Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 6 Adaptive Behavior Figure 2. Flowcharts depicting adult-agent (a) and juvenile-agent (b) decision rules for each time step. At the beginning of a time step, all agents increase one unit in age. When adult agents reach the end of their lifespan or have no resources, they die. When juvenile agents reach the end of their juvenile period, they become adults. If a juvenile agent’s parent(s) dies before it becomes an adult, the juvenile agent dies. Table 1. Fixed parameters. Description Parameter Values and/or descriptions Payoffs Base life span Standard deviation Reproduction resource threshold Resource cap Parental investment C T, D S L SDL RT Rc pI Parental care pc Mutation rate Grid size Aggregation threshold Search radius r X×Y AT SR Juvenile development D 2 – Cooperator payoff 1 – Temptation and defector payoffs 0 – Sucker’s payoff 250 – Time steps 30 – The standard deviation for L 100 – The resource level required for an agent to reproduce. 300 – Maximum resources an agent can have 0.1 – The percentage of parent resources initially invested when producing an offspring 0.01– The percentage of parent resources a juvenile offspring receives each time step 0.01 50 × 50 – Spatial dimensions in cells 1 – The minimum number of agents an agent to contact attempts to contact 2 – The distance in cells, an agent searches for other agents when attempting to aggregate 50 – Rounds of play Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 7 Table 2. Parameters sweeps. Description Parameter Values Learning rate Population size Reproduction For sexual reproductive cases: Allelic interaction α Nmax 0.04, 0.06, 0.08, 0.1, 0.2 250, 500, 1000 Sexual, asexual Recessive, additive, dominant Biparental, uniparental Parental care type Table 3. Initial conditions. Description Parameter Values Average starting age Standard deviation for starting age Average starting resources Standard deviation for starting resources Initial population size SA SDSA SR SDSR N 150 50 75 30 pmax/2 reproductive conditions, we systematically varied the rate of learning and maximum population size (or density–holding grid size constant) for a total of 5 3 3=15 conditions (Table 2). For each condition, we ran 100 runs of 100,000 time steps each. 3.3 Population structure control conditions Because small groups of agents formed over time due to active aggregation and limited dispersal, group selection emerged in these simulations (Figure 3). To eliminate the effects of group selection; a second set of control simulations was also conducted. These simulations were identical to the simulation conditions in the previous section except that on each round of play, all agents of the same developmental stage randomly swapped locations with each other. This eliminated spatial associations among agents (Figure 3). These simulations allowed us to assess the role of population structure in the evolution of social play. Figure 3. An illustration of initial conditions (a, b) and agent swapping (c, d). Initially, agents are non-cooperators (solid red), lack play alleles, and are randomly distributed in unique locations in space (a). Early in a simulation, agents aggregate into groups of varying sizes (adults are solid and juveniles are open circles) and a few juvenile agents have the social-play gene by mutation (blue circle; b). Later in the simulation, small clusters of cooperators (shaded blue according to their probability of cooperating) of varying sizes with cooperators, non-cooperators, and offspring with and without the play gene emerge (c). To illustrate location swapping, on the next round, the agents in (c) randomly swap locations with agents in their developmental class (d). the recessive case, if both alleles are cooperative, agents cooperate with probability 1.0; otherwise, they cooperate with probability 0.0. For the additive case, agents with one cooperative allele cooperate with probability of 0.5, agents with two cooperative alleles cooperate with probability 1.0, and agents with no cooperative alleles cooperate with probability 0.0. For the dominant case, agents with at least one allele always cooperate with probability 1.0; otherwise, they cooperate with probability 0.0. 4. Results and discussion 3.4 Cooperation control conditions 4.1 Social play population structure In a second set of control conditions, we systematically investigated the evolution of cooperation without social play. For both sexual and asexual reproduction, cooperation alleles were inherited in the same manner as play alleles. We ran 18 conditions corresponding to the sexual reproduction conditions and three conditions corresponding to the asexual reproduction conditions in Section 3.2 (varying the rate of learning was not applicable here). For sexual reproduction, we defined the probabilities for cooperation in the SH as follows. For Social play evolved by facilitating the learning of adult cooperation in population structures characterized by the formation of spatial clusters or groups of agents, and evolved even when the rate of learning was relatively low (Figure 5). In contrast, in the cooperation control conditions (where agents cooperated only if they inherited a cooperation gene), cooperation evolved infrequently in most conditions and only evolved in slightly more than 50% of the simulations in the sexually reproducing dominant allele condition with low Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 8 Adaptive Behavior Figure 4. A succession of snapshots of the evolution of social play and adult cooperation over 100,000 rounds for a small population size (Nmax=250) under the sexual, uniparental care condition. Juveniles with a play gene are illustrated with open blue circles and juveniles without a play gene are illustrated with red open circles. Adults that have learned to cooperate are represented by shaded-blue circles (i.e., shaded blue according to their probability of cooperating) and non-cooperators represented by solid red circles. In the early stages of evolution, the social play gene (and learned cooperation) increase in small regions of space. When the frequency of cooperation in these small regions exceeds the fixed-point p=0.5, cooperators can then expand as illustrated in the bottom two rows. population density (Figure 6). In general, cooperation without social play evolved much less frequently and at much lower frequencies than cooperation facilitated by social play (cf. Figures 5 and 6). For cooperation to evolve in the SH, the frequency of cooperators must exceed the fixed-point frequency of 0.5 (i.e., ESS zone for the evolution of cooperation). Because populations began with no cooperators and low mutation rates, the frequency of cooperators was well below the unstable fixed point making it effectively impossible for cooperation to increase in panmictic populations, which is exactly what we found in the swapping control conditions (Figure 7). With population structure (i.e., agents actively aggregate and offspring locally disperse), however, the frequency of cooperators varied over local regions of space (Figure 4). Occasionally, the frequency of cooperators in a region of space exceeded the fixed point and gradually expanded throughout the whole population (Figure 4). This was most likely to occur for relatively small population size, sexual reproduction with dominance, and uniparental care (Figure 5). Conditions with smaller maximum population size (or density) tended to evolve cooperation more often because clusters of agents were more likely to be isolated from one another, which promoted group selection (Figures 5 and 6). By adding social play, adult cooperation dramatically increased (Figure 5). The explanation of this effect is that the negative effect of cooperating in a population of non-cooperators is masked at low frequencies by social play. When a play gene exists at low frequencies in a population, the frequency of adult cooperators is close to zero because the likelihood that two or more juveniles with the play gene are next to each other is low. In a structured population characterized by spatially isolated groups, this can facilitate (by genetic drift) increases in the frequency of a social play gene within small regions of space (Figure 4). As mutation and drift increase the frequency of play genes within Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 9 Figure 5. The frequency of simulations for which the average probability of cooperating was greater than 0.5 (i.e., the evolutionarily stable strategies zone for cooperation). The learning rate, α, was varied from 0.04 (weak) to 0.2 (strong) as illustrated on the horizontal axes. The first column (a, c, e) contains graphs of parameter sweeps for the sexual biparental care conditions. The second column (b, d, f) contains graphs of parameter sweeps for sexual uniparental care conditions and the asexual condition. Each row contains graphical results for different population sizes: the first row (a, b) is for a population size of 250, the second row (c, d) is for a population size of 500 and the third row (e, f) is for a population size of 1000. these regions, the frequency of juveniles with the play gene can become sufficiently high so that agents begin to learn to cooperate. Because agents that play together as juveniles tend to stay in the same place as adults, the likelihood of one cooperator finding others like it are relatively high in these regions of space. Thus, small regions emerge in which cooperators do relatively well when their frequency is greater than the fixed-point value of p=0.5 and thereby out reproducing regions with non-cooperators, and quickly spread throughout a population (Figure 4). 4.2 Sexual reproduction and parental care Social play was more likely to evolve under the dominant and additive sexual reproductive conditions than Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 10 Adaptive Behavior Figure 6. The frequency of simulations for which the average probability of cooperating was greater than 0.05 (i.e., the evolutionarily stable strategies zone for cooperation) by population size in which agents did not learn to play but instead but instead evolved cooperation via a cooperation gene. On the left (a) are the sexual biparental care conditions. On the right (b) are the sexual uniparental care conditions and the asexual case. the asexual and recessive conditions (Figure 5). Sexually reproducing agents with a recessive play gene did the worst because they required two copies of the play gene for the expression of social play. For both the dominant and additive cases of sexual reproduction, cooperative agents did quite well because only one gene was required for full or partial expression. A freeloading effect emerged for both the dominant and additive cases. Freeloading is straightforwardly explained in the uniparental care condition. In this condition, one agent assumes the role of the care-giving parent (CG) and provides all of the parental care, while the other agent only provides a genetic contribution. In a population with a low frequency of cooperators, an offspring with a cooperative CG parent would be at a disadvantage to an offspring with a non-cooperative CG parent. However, if the genetic-contribution-only (GCO) parent happens to be a cooperator and the CG parent is a non-cooperator, then the GCO parent may pass on the social play gene to their offspring—and the offspring will not incur any costs associated with having a cooperative GCO parent in a predominately noncooperative population. Furthermore, the cooperative GCO parent will not be burdened with providing parental care, and hence will have the resources necessary to reproduce again shortly afterwards. Thus, it is possible for a cooperator parent to reproduce two or more juvenile offspring with a play allele in quick succession, thereby increasing the frequency of play allele offspring in a particular group to the levels needed for cooperation and social play to take off. In the case of biparental care, both parents contribute equal resources. If an offspring with the play allele has a cooperator and a non-cooperator as parents, it will benefit from the noncooperator parent when adult cooperation is low, but not as much as in the uniparental case. A cooperative parent may also be more likely to reproduce multiple offspring with a social play allele in a short time period, though it will not be able to reproduce as quickly as in the uniparental case, due to the required parental care. Figure 5 illustrates that both dominant and additive conditions did very well under both biparental and uniparental care. The biparental dominant and additive conditions did slightly better than the same uniparental conditions only for low learning rates and high population size (or density). The opposite was true for the asexual and recessive conditions. In the asexual condition, there was no freeloading effect as offspring with a cooperator parent only received resources from that cooperator parent and therefore incurred a significant cost when the population was predominantly non-cooperative. Cooperative parents were also much less likely to have the opportunity to reproduce multiple offspring consecutively. This could explain why asexual cases did not often evolve cooperation (Figures 5–7). Finally, the freeloading effect is frequency dependent. Freeloading is only beneficial to the evolution of cooperation and social play when a population is predominantly non-cooperative and adult cooperators receive no payoff the majority of the time. Once cooperation has spread throughout a population, cooperators obtain the cooperative payoff the majority of the time and there is no longer a benefit to freeloading on noncooperative parents. Freeloading is a two way street when cooperators are common; it becomes detrimental to cooperation because offspring without the play allele can freeload on cooperative parents. 5. General discussion We found that social play gains delayed adaptive benefit by facilitating the learning of adult cooperation. For adults playing the SH, cooperation can evolve from Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 11 Figure 7. The frequency of simulations for which the average probability of cooperating was greater than 0.5 (i.e., the evolutionarily stable strategies zone for cooperation) by learning rate, α, for agent location swapping, which eliminated group effects. This graph shows the results for a population size of 250. For all other population sizes investigated, cooperation frequencies were lower than shown here. very low initial frequencies when population structure is characterized by the formation of spatial clusters of agents. Social play as a necessary requirement for cooperation, in our model, allows juveniles to learn to cooperate only when there are other juveniles present that engage in play. Because agents remain at the same location so long as they are in contact with at least one other agent, it is likely that an agent that learned to cooperate by engaging in social play will remain next to agents that also learned to cooperate when they enter adulthood. Thus, agents that engage in social play as juveniles are more likely to benefit from higher cooperative payoffs as adults, resulting in selection for social play. Our theoretical results support the hypothesis that social play could evolve by promoting the learning of cooperation when starting with populations of entirely non-cooperative agents. Our model and results provide a novel solution to the problem of the evolution of cooperation, a notoriously difficult problem to explain. For the SH, for a wide range of parameter values, cooperation cannot invade a population of non-cooperators. For cooperative games such as the prisoner’s dilemma, several mechanisms have been found that support the evolution of cooperation including retaliatory strategies such as tit-for-tat (Axelrod & Hamilton, 1981), walking away from defectors (Aktipis, 2004; Smaldino & Schank, 2012), and spatial aggregation (Smaldino, Schank, & McElreath, 2013; Traulsen & Nowak, 2006). We have shown that social play in conjunction with spatial aggregation could be a very powerful mechanism for the evolution of cooperation. Interestingly, social play only robustly evolves by facilitating adult cooperation under sexual reproduction. Assuming a single play gene that is either dominant or additive for social play resulted in the evolution of social play even when learning was very gradual and required a number of bouts of social play to learn cooperation. We argue that sex allows the play gene to freeload on non-cooperator parents when the play gene is relatively rare. Freeloading frees offspring, to some extent, from selection on parents that have learned to cooperate when the frequency of adult cooperators is low. This allows the play gene to increase by drift in small clusters of agents. By chance, clusters occasionally emerge with relative high frequencies of the play gene; members of these clusters engage in social play and subsequently cooperation. Once a group of agents has over 50% cooperators (i.e., for the payoff structure used in these simulations, the group of agents enter the ESS zone for cooperation), it has a fitness advantage over groups of non-cooperators and cooperation is locally favored. The evolutionary process by which this occurs is similar to Wright’s (1932) shifting balance theory. In our model, we assumed that juvenile agents engaged in play and that social play was one form of play with no additional costs. Social play may introduce costs in addition to the costs of other forms of play. As outlined in the introduction, these potential fitness costs can take several forms. Increased predation (Blumstein, 1998; Harcourt, 1991), injuries (Berger, 1980; Douglas-Hamilton & Douglas-Hamilton, 1975), energy costs (Martin, 1984; Miller & Byers, 1991), and foraging costs (Caro, 1995). Each of these costs can entered into a model such as ours in different ways. We have begun to investigate different forms of costs to social play. For example, in simulations from a new model we are analyzing, which includes the cost of social play as increased risk of predation, we have found that social play still evolve for a wide-range of parameters reported above when predation risk is as high as 10% during the juvenile developmental period. Future research will investigate how the potential costs of social play affect its evolution. Our results suggest that the frequency and complexity of social play in juveniles should correlate with the frequency and complexity of adult cooperation found in different mammalian species. There is evidence that juvenile animals learn to cooperate and behave fairly through social play (Bauer & Smuts, 2007; Bekoff, 1977, 2001; Brosnan & de Waal, 2012; Dugatkin & Bekoff, 2003; Pellis et al., 2009; Watson & Croft, 1996). There is also evidence indicating that social play in early development is important for social skills in adulthood (Einon et al., 1978; Hol et al., 1999; Pellegrini, 1992; Pellis & Pellis, 2007). However, further research is needed to determine whether the social skills learned Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 12 Adaptive Behavior during play are used in adult cooperative behavior and whether these skills translate into adaptive benefits. Our model also predicts that the evolutionary context in which cooperation and social play evolved consisted of populations of small isolated groups, rather than large, well-mixed populations. This prediction can only be assessed once we have made progress on the former question. Although we have shown that social play could evolve by facilitating adult cooperation, cooperation is more complicated and dynamic than mere coordination of behavior for a cooperative goal (e.g., see MacLennan & Burghardt, 1993, for a general discussion of the evolution of cooperation in the context of synthetic ethology). We modeled a simple coordination scenario in which individuals must coordinate their behavior to achieve a resource. In addition to achieving a cooperative gain, the resource must then be divided, which introduces the possibility of agents fighting over the resource or one individual behaving unfairly, taking the cooperative gain for itself (defecting). This suggests an important direction for future research. Could juvenile agents learn social skills during social play for dealing with defectors? Models could be investigated in which there is initial variation in the degree to which juvenile agents are likely to engage in fair play and they subsequently learn to play fairly or not as a function of their bouts of social play. We know that animals may learn how to play fairly and solve conflicts during play (Dugatkin & Bekoff, 2003). If, for example, fair agents learn to avoid unfair agents (Brosnan & de Waal, 2012), they may have a successful counter strategy for defectors. Acknowledgments This work was inspired by participants in the working group on Play, Evolution, and Sociality sponsored by National Institute for Mathematical and Biological Synthesis, an Institute sponsored by the National Science Foundation through NSF Award #DBI-1300426 with support from The University of Tennessee, Knoxville. References Aktipis, C. A. (2004). Know when to walk away: Contingent movement and the evolution of cooperation. Journal of Theoretical Biology, 231, 249–260. Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211(4489), 1390–1396. Baldwin, J. D., & Baldwin, J. I. (1974). Exploration and social play in squirrel monkeys (Saimiri). American Zoologist, 14(1), 303–315. Bauer, E. B., & Smuts, B. B. (2007). Cooperation and competition during dyadic play in domestic dogs, Canis familiaris. Animal Behaviour, 73(3), 489–499. Bekoff, M. (1974). Social play and play-soliciting by infant canids. American Zoologist, 14(1), 323–340. Bekoff, M. (1995). Play signals as punctuation: The structure of social play in canids. Behaviour, 132, 419–429. Bekoff, M. (2001). Social play behavior: Cooperation, fairness, trust, and the evolution of morality. Journal of Consciousness Studies, 8(2), 81–90. Bekoff, M., & Allen, C. (1998). Intentional communication and social play: How and why animals negotiate and agree to play. In M. Bekoff, & J.A. Byers (Eds.), Animal play: Evolutionary, comparative, and ecological perspectives. Cambridge, UK: Cambridge University Press. Bekoff, M., & Pierce, J. (2009). Wild justice: The moral lives of animals. Chicago, IL: University of Chicago Press. Berger, J. (1980). The ecology, structure and functions of social play in bighorn sheep (Ovis canadensis). Journal of Zoology, 192(4), 531–542. Blumstein, D. T. (1998). Quantifying predation risk for refuging animals: A case study with golden marmots. Ethology, 104(6), 501–516. Brosnan, S. F., & de Waal, F. B. (2012). Fairness in animals: Where to from here?Social Justice Research, 25(3), 336–351. Burghardt, G. M. (1982). Comparison matters: Curiosity, bears, surplus energy, and why reptiles do not play. Behavioral and Brain Sciences, 5(01), 159–160. Caro, T. M. (1980). Effects of the mother, object play, and adult experience on predation in cats. Behavioral and Neural Biology, 29(1), 29–51. Caro, T. M. (1995). Short-term costs and correlates of play in cheetahs. Animal Behaviour, 49(2), 333–345. Douglas-Hamilton, I., & Douglas-Hamilton, O. (1975). Among the elephants. London, UK: Collins. Dugatkin, L. A., & Bekoff, M. (2003). Play and the evolution of fairness: A game theory model. Behavioural Processes, 60, 209–214. Easton, D. M., & Hirsch, H. R. (2008). For prediction of elder survival by a Gompertz curve, number dead is preferable to number alive. Age, 30(4), 311–317. Einon, D. F., Morgan, M. J., & Kibbler, C. C. (1978). Brief periods of socialization and later behavior in the rat. Developmental Psychobiology, 11(3), 213–225. Fagen, R. M. (1981). Animal Play Behavior. Oxford, UK: Oxford University Press. Gomendio, M. (1988). The development of different types of play in gazelles: Implications for the nature and functions of play. Animal Behaviour, 36(3), 825–836. Graham, K. L., & Burkhardt, G.M. (2010). Current perspectives on the biological study of play: Signs of progress. The Quarterly Review of Biology, 85(4), 393–418. Harcourt, R. (1991). Survivorship costs of play in the South American fur seal. Animal Behaviour, 42(3), 509–511. Hayaki, H. (1985). Social play of juvenile and adolescent chimpanzees in the Mahale Mountains National Park, Tanzania. Primates, 26(4), 343–360. Hol, T., Van den Berg, C.L., Van Ree, J. M., & Spruijt, B. M. (1999). Isolation during the play period in infancy decreases adult social interactions in rats. Behavioural Brain Research, 100, 91–97. Lee, P. C. (1983). Play as a means for developing relationships. Primate Social Relationships, 82–89. Luke, S., Cioffi-Revilla, C., Panait, L., Sullivan, K., & Balan, G. (2005). MASON: A multiagent simulation environment. Simulation, 81(7), 517–527. Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 Durand and Schank 13 Mackey, A. D., Makecha, R. N., & Kuczaj, S. A. (2014). The development of social play in bottlenose dolphins (Tursiops truncatus). Animal Behavior and Cognition, 1(1), 19–35. MacLennan, B. J., & Burghardt, G. M. (1993). Synthetic ethology and the evolution of cooperative communication. Adaptive Behavior, 2(2), 161–188. Martin, P. (1984). The time and energy costs of play behaviour in the cat. Zeitschrift fur Tierpsychologie., 64(3–4), 298–312. Miller, M. N., & Byers, J. A. (1991). Energetic cost of locomotor play in pronghorn fawns. Animal Behaviour, 41(6), 1007–1013. Nunes, S., Muecke, E. M., Sanchez, Z., Hoffmeier, R. R., & Lancaster, L. T. (2004). Play behavior and motor development in juvenile Belding’s ground squirrels (Spermophilus beldingi). Behavioral Ecology and Sociobiology, 56(2), 97–105. Pellegrini, A. D. (1992). Rough-and-tumble play and social problem solving flexibility. Creativity Research Journal, 5(1), 13–26. Pellegrini, A. D., Dupuis, D., & Smith, P. K. (2007). Play in evolution and development. Developmental Review, 27, 261–276. Pellis, S. M., & Pellis, V. C. (2007). Rough-and-tumble play and the development of the social brain. Current Directions in Psychological Science, 16(2), 95–98. Pellis, S. M., Field, E. F., & Whishaw, I. Q. (1999). The development of a sex-differentiated defensive motor pattern in rats: A possible role for juvenile experience. Developmental Psychobiology, 35(2), 156–164. Pellis, S. M., Pellis, V. C., & Reindhart, C. J. (2010). The evolution of social play. In C. M. Worthman, P. M. Plotsky, D. S. Schechter, & C. A. Cummings (Eds.), Formative experiences: The interaction of caregiving, culture, and developmental psychobiology (pp. 404–431). Cambridge, UK: Cambridge University Press Power, T. G. (2000), Play and exploration in children and animals. Hillsdale, NJ: Lawrence Erlbaum Associates. Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A.H. Black, & W.F. Prokasy (Eds.), Classical conditioning: Current research and theory. New York: Appleton-Century-Crofts. Rousseau, J. J. (1992). Discourse on the origin of inequality. Indianapolis, IN: Hackett Publishing. Schank, J. J., Smaldino, P. E., & Miller, M. L. (2015). Evolution of fairness in the dictator game by multilevel selection. Journal of Theoretical Biology, 382: 64–73. Sharpe, L. L. (2005). Play fighting does not affect subsequent fighting success in wild meerkats. Animal Behaviour, 69(5), 1023–1029. Skyrms, B. (2004). The stag hunt and the evolution of social structure. Cambridge, UK: Cambridge University Press. Smaldino, P. E., & Schank, J. C. (2012). Movement patterns, social dynamics, and the evolution of cooperation. Theoretical Population Biology, 82, 48–58. Smaldino, P. E., Schank, J. C., & McElreath, R. (2013). Increased costs of cooperation help cooperators in the long run. The American Naturalist, 181(4), 451–463. Sussman, R. W., Garber, P. A., & Cheverud, J. M. (2005). Importance of co-operation and affiliation in the evolution of primate sociality. American Journal of Physical Anthropology, 128(1), 84–97. Thor, D. H., & Holloway, W. R. (1985). Social play in juvenile rats: A decade of methodological and experimental research. Neuroscience & Biobehavioral Reviews, 8(4), 455–464. Traulsen, A., & Nowak, M. A. (2006). Evolution of cooperation by multilevel selection. Proceedings of the National Academy of Sciences of the United States of America, 18, 10952–10955. Van den Berg, C.L., Hol, T., Van Ree, J. M., Spruijt, B. M., Everts, H., & Koolhaas, J. M. (1999). Play is indispensable for the adequate development of coping with social challenges in the rat. Developmental Psychobiology, 34(2), 129–138. Vidya, T. N. C., & Sukumar, R. (2005). Social and reproductive behaviour in elephants. Current Science, 89(7), 1200–1207. Watson, D. M. (1998). Kangaroos at play: Play behaviour in the Macropodoidea. In M. Bekoff, & J. A. Byers (Eds.,) Animal play: Evolutionary, comparative, and ecological perspectives (pp. 61–95). Cambridge, UK: Cambridge University Press. Watson, D. M., & Croft, D. B. (1996). Age-related differences in playfighting strategies of captive male red-necked wallabies. Ethology, 102(2), 336–346. Wright, S. (1932). The roles of mutation, inbreeding, crossbreeding and selection in evolution. Proceedings of the Sixth International Congress of Genetics 1, 356–366. Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016 14 Adaptive Behavior About the Authors Sabine Durand is a graduate student studying psychology at the University of California, Davis. She received her B.A. in Psychology and Mathematics at University of California, Santa Cruz in 2011. She is interested in complex group and social behaviors, particularly cooperation. Jeffrey C Schank is a professor of psychology at the University of California, Davis. He received his PhD in the conceptual foundations of science from the University of Chicago in 1991. His primary research tool is agent-based modeling, which he applies to the study of social behavior, its organization, development, and evolution. Downloaded from adb.sagepub.com at UNIV CALIFORNIA DAVIS on March 3, 2016
© Copyright 2026 Paperzz