Hum Nat DOI 10.1007/s12110-014-9199-y

The Two Sides of Warfare An Extended Model of Altruistic Behavior in Ancestral Human Intergroup Conflict Hannes Rusch

# Springer Science+Business Media New York 2014

Abstract Building on and partially refining previous theoretical work, this paper presents an extended simulation model of ancestral warfare. This model (1) disentangles attack and defense, (2) tries to differentiate more strictly between selfish and altruistic efforts during war, (3) incorporates risk aversion and deterrence, and (4) pays special attention to the role of brutality. Modeling refinements and simulation results yield a differentiated picture of possible evolutionary dynamics. The main observations are: (a) Altruism in this model is more likely to evolve for defenses than for attacks. (b) Risk aversion, deterrence, and the interplay of migration levels and brutality can change evolutionary dynamics substantially. (c) Unexpectedly, one occasional simulation outcome is a dynamically stable state of “tolerated intergroup theft,” raising the question as to whether corresponding patterns also exist in real intergroup conflicts. Finally, possible implications for theories of the coevolution of bellicosity and altruism in humans are discussed. Keywords Intergroup conflict . Cooperation . Public goods . Altruism . Warfare Although disenchanting, it is an intriguing and recently much discussed idea that violent intergroup conflict (“warfare”) might have been an important driver in the evolution of human social behavior (e.g., Bowles 2009; Choi and Bowles 2007; Ginges and Atran 2011; Gneezy and Fessler 2011; Halevy et al. 2012; Konrad and Morath 2012; Lehmann and Feldman 2008; Saaksvuori et al. 2011; Smirnov et al. 2007). Resemblances between chimpanzee and human intergroup conflict (Goodall 1986; Mason and Mendoza 1993; Wrangham and Glowacki 2012) as well as the archaeological record and anthropological observations (Gat 2009; Keeley 1996; Kelly 2005) make it plausible that warfare had an influence on human evolution from very early on. Electronic supplementary material The online version of this article (doi:10.1007/s12110-014-9199-y) contains supplementary material, which is available to authorized users. H. Rusch (*) Behavioral and Institutional Economics, JLU Giessen, Licher Strasse 66, 35394 Giessen, Germany e-mail: [email protected]

Hum Nat

Thus, behavioral and psychological adaptations to the problems posed by violent intergroup conflict are not at all unlikely (McDonald et al. 2012; van Vugt et al. 2007). It has even been hypothesized repeatedly (Bowles 2009; Choi and Bowles 2007; Ginges and Atran 2011; Halevy et al. 2012; Konrad and Morath 2012; Lehmann and Feldman 2008; Smirnov et al. 2007) that in human evolutionary history there might have been a coevolution of the readiness to engage in violent intergroup conflict (“bellicosity”) and group-beneficial altruistic behaviors—i.e., the readiness to incur fitness costs, even up to sacrificing one’s own life, in order to help one’s in-group, or to engage in costly punishment of unwilling in-group members (Gneezy and Fessler 2011; Mathew and Boyd 2011; Saaksvuori et al. 2011). For altruism and bellicosity to coevolve, however, a number of prerequisites have to be met (see, e.g., Bowles 2009). Among them are (1) high genetic and/or cultural heritability of both bellicosity and altruism, (2) rather low levels of genetic and/or cultural exchange between losing and winning groups, (3) a positive influence of victory in war on the reproductive and/or social success of the victors and (4) a high frequency and intensity of wars, enabling war to become a selective force over evolutionary periods of time. Each of these assumptions deserves a thorough investigation of its own. Here, however, I focus on the most important assumption, namely that (5) warfare itself is an undertaking in which altruism plays a significant role. Building on previous theoretical works, I develop an extended model of ancestral human intergroup conflict. The model tries to accommodate a number of empirically established characteristics of intergroup conflict that have not yet been the focus of theoretical approaches. This extended model thus yields a somewhat less abstract account of the incentives for and in warfare and the evolutionary trajectories they might entail. The psychological and behavioral adaptations for violent intergroup conflict that are the focus of this paper may have evolved as early as in the common ancestor of humans (Homo) and chimpanzees and bonobos (Pan). Chimpanzee intergroup conflict still resembles human warfare (Wrangham and Glowacki 2012), whereas bonobos do not show such high levels of violence between groups (Boehm 2012; Wrangham 1999). The standard model for the following considerations, however, will be that of small human hunter-gatherer groups since this form of social organization and subsistence likely has governed most of our specifically human evolutionary past (Hill et al. 2011; Keeley 1996). Furthermore, warfare has almost always been an exclusively male occupation under ancestral conditions (Pratto et al. 2006; Tooby and Cosmides 2010). Thus, the following considerations apply mostly to the evolution of the motivational psychology of males for violent intergroup conflict (McDonald et al. 2012).

Modeling Ancestral Intergroup Conflict Is War an n-person Prisoner’s Dilemma? Many previous models frequently either explicitly claim or simply assume that the game theoretical structure of warfare is symmetrical. Some researchers use the wellstudied paradigm of the n-person Prisoner’s Dilemma (the public goods game: PGG) to model the cost/benefit structure of ancestral warfare (e.g., Bowles 2009). This

Hum Nat

assumption implies, however, that the strategic structure of warfare for all individuals on both sides should essentially have a form as given in Table 1. Although this assumption facilitates mathematical analyses, it is questionable whether this view sufficiently captures the strategic structure of warfare. A distinction between the attackers’ and the defending party’s perspective is hardly dispensable, and non-participation does not always seem to be the best choice, as the following considerations will show. Let us briefly review the goods produced in attack (i.e., loot and status, deterrence, and territory) and defense (i.e., safety and deterrence) and ask to which of them the PGG logic is likely to apply in reality, and to which it is not. Attack Loot and Status Imagine that your group is organizing a raid against another group, intending to rob valuable resources such as food, tools, livestock, or—especially important—women (Gat 2009; Kohler and Kramer Turner 2006). For such a raid to be an n-person Prisoner’s Dilemma, non-participation must be the strictly dominant strategy—in other words, it must always be the best choice for the individual. Thus, non-participation must result in a higher gain for you than participating in the attack would. For successful raids this would imply that the raiders return home and give you, the non-participator, a fair share of the loot they captured. This fair division of all goods produced through warfare is a crucial assumption implicitly made by the PGG approach to modeling warfare. It seems quite questionable, though, that the fitnessrelevant benefits of successfully engaging in raids are actually shared with nonparticipators. This point particularly applies to individual status increases (see, e.g., Chagnon 1988) and the additional access to women gained through warfare, which, while being highly fitness relevant, is unlikely to be shared even within the group of combatants (Gat 2009). I will return to this point in the “Discussion.” Deterrence When successfully raiding another group, your fellow group members might produce deterrence, which they cannot deny you. Deterrence is thus a good candidate for a public good produced through war. By not participating, however, you would still lose in terms of material resources and status (Gat 2009)—and likely even reproductive success (Chagnon 1988; Kohler and Kramer Turner 2006)—relative to the successful raiders in your group. Apart from the deterrence produced, they will attain concrete additional resources and you will not. Thus, while possessing the basic features of a public good, deterrence might be better described as a by-product (West et al. 2007) of individually beneficial raiding, not as a strategic goal of war. Finally, Table 1 PGG structure of war All other individuals… Individual …

Participate

Do not participate

Participates

Good result (costly victory)

Worst result (e.g., death)

Does not participate

Best result (costless victory)

Bad result (e.g., group dispersal)

Individual’s payoffs, best replies marked in bold

Hum Nat

while producing the short-term benefit of increased deterrence, raids likely also induce an increased risk of subsequent retaliatory strikes by the raided group (Beckerman et al. 2009), which might even level the earlier benefits of deterrence. Territory Thus, among the potential benefits of successful raiding—such as the loot, including women; increases in individual status; and deterrence and territorial expansion—only the last seems to pose a good candidate for being a public good intentionally produced through warfare. However, describing warfare as a means for territorial expansion which thereafter benefits the entire group might eventually be putting the cart before the horse: Taking into account that more prosperous groups (i.e., those groups holding the ecologically richer territories in the first place) are likely to be larger, and thus likely to have more potential warriors, it might also be the case that territorial expansion by means of warfare is a consequence of group prosperity, and not its cause. The finding reported by Wrangham and Glowacki (2012) that territorial expansion was rarely stated as a motivation for attacks in emic accounts of intergroup conflict, although warfare frequently does result in such, seems to render support for this view. Furthermore, Glowacki and Wrangham (2013) recently established that, when territorial gains are excluded, prospects of individual benefits attainable through raiding are positively correlated with the frequency of violent intergroup conflicts in small-scale societies. Additionally, as Kelly (2005) points out, there is an important trade-off problem to take into account regarding aggressive territory expansion: ongoing intergroup conflict over territory produces buffer zones which can no longer be used safely by either party. This means that, as long as no group is superior enough to capture quite large amounts of land, remaining peaceful can effectively be the best means of maximizing accessible territory.

Defense Safety Now imagine, on the other hand, your group being attacked by a hostile group. According to a PGG view of war, in this situation your best strategic choice would always be not to defend yourself, e.g., by trying to flee, no matter what your fellow group members—including your family—do (remember: non-participation in war must be a strictly dominant strategy). This theoretical claim again seems counterfactual. In this case, depending on the brutality and the resoluteness of the attackers, your personal survival, your reproductive success, and the survival of many of your kin can depend on your group’s ability to mobilize a force strong enough to fend the attackers off from raiding or destroying your resources, including the women and children of your group. In game theoretical terms, a threshold PGG is played in this situation. In order for cooperation—i.e., participation in the defense of the group—to evolve in threshold PGGs, no altruism is necessary under a range of circumstances (Bach et al. 2006). Rather, participation frequently is the fitness-maximizing strategy in this situation. Thus, the choice faced by individuals in defending groups is likely not as straightforward as in simple PGGs. Imagine, e.g., that your group is lacking only one more defender in order to reach the threshold required to fend off an incoming attack. If you choose to desert, chances are high that you and/or your kin will lose a substantial amount of fitness-relevant resources. If you choose to participate in the defense, however, you might die. Obviously, your decision now depends on the comparison

Hum Nat

of these two risks. Even if the threshold for a successful defense is already reached, if we assume that success in war is not completely determined by fighting strength but is to some degree also influenced by chance events, it might still be that supporting the defense of your group has a higher expected return than deserting, especially because the risk of dying for the individual defender is likely to decrease with the number of active defenders. Deterrence As in attack, given the potential direct individual benefits of successful defense, it seems plausible to assume that the deterrence produced by a strong group defense, while actually being a public good, is not a primary objective of the defensive effort but rather its by-product. If this reasoning is sound, strategic considerations significantly differ for attack and defense and there are no dominant strategies concerning participation in war: For attacks it is better to participate if the attack is likely to be successful, better not to participate if the attack is likely to fail. For defense it might be better not to take part in a successful defense, but potentially devastating if the defense fails, raising the challenging problem of assessing how much to invest in order to reach the required threshold. For both types of warfare, attack and defense, decisions crucially depend on the expected chance of success of the undertaking, which itself, in both cases, depends on the expected behavior of fellow group members and that of the other group’s members, making simple dominant strategies impossible (for related discussions see Durham 1976; Gat 2009; Kelly 2005; Tooby and Cosmides 2010). In game theoretical terms: Given the above considerations, attack is probably best characterized as a “threshold club good game” which produces some less important public goods as by-products, whereas the strategic structure of defense is captured by a threshold public goods game. The model presented in the following is set up accordingly. An Alternative Payoff Structure To overcome these weaknesses of the PGG approach, we need to define a more appropriate set of payoff functions. First, we should account for the different meanings of victory for the two parties at war: successful attackers gain fitness-enhancing resources (i.e., capture loot, women, or territory), while successful defenders merely do not lose resources. Since attackers usually leave their territory when going on a raid and withdraw when they are not victorious, it is unlikely that successful defenders can capture more than some weapons or other personal items from the defeated attackers (Kelly 2005). Second, we should try to disentangle “selfish” (i.e., individually fitness maximizing), from “altruistic” (i.e., individually costly but group beneficial) investments in war efforts. Since the individual itself and his kin are part of the group (Hill et al. 2011), group-beneficial behavior per se is not “pure altruism.” A detailed discussion of the implications of this quite simple fact for the concept of altruism can be found, for example, in Kerr and Godfrey-Smith (2002), who introduce the terms “weak” and “strong” altruism to analyze this situation. Weak altruism describes an individual’s investment in a group effort which, while also benefitting the other group members, including those who do not take part in the group’s collective action, results in a higher

Hum Nat

payoff to the individual than choosing not to invest. This payoff structure can, for example, be induced by the kinship composition of the group (see Lehmann et al. 2007; Rusch 2014). Strong altruism, on the other hand, describes an investment in a common effort which yields lower returns than free-riding. To accommodate this important distinction, two sorts of investments are introduced to the analysis of intergroup conflict: “selfish” (or “weakly altruistic” in Kerr and Godfrey-Smith’s terminology) investments that benefit the group but also bring about a relative increase in the share that the investing individual will receive of those returns produced through warfare which are dividable, and (“strongly”) “altruistic” investments that do not result in a higher share of the benefit for the individual while still increasing the aggregate returns for the group. At this point, it is already clear to the sharp eye that these distinctions limit the scope of altruism mostly to defense: The goods produced in defense are non-excludable and non-dividable, whereas the most important goods produced in attacks are dividable and can be exclusively owned. This, admittedly, is close to self-evident, but it has been disregarded by many previous models of the evolution of bellicosity nevertheless. The distinction between “selfish” and “altruistic” investments in defense is thus dispensable at present, but will be important later. As a third requirement the payoff for victorious attackers and the loss for defeated defenders should be proportional to the superiority of the attack instead of being “all-or-nothing.” This claim is motivated by the observed characteristics of hunter-gatherer intergroup conflict. Although probably existent (Beckerman et al. 2009), wars of extermination resulting in the annihilation of defenders and “total victory” of the attackers presumably are not a representative form of hunter-gatherer warfare (Fry and Söderberg 2013; Gat 2009; Keeley 1996; Kelly 2005). Rather, it has often been observed that groups of attackers go on raids and return with as much booty as they can capture before personal risk becomes too high (Mathew and Boyd 2011; Pitman 2011; Wrangham and Glowacki 2012). By mobilizing a strong defense, the attacked can thus reduce their loss, i.e., the payoffs to the attackers. Model Structure: Benefits and Costs The simulation model used here meets all of the aforementioned demands. For the sake of analytical feasibility, and to be comparable with earlier models, it uses g groups with a fixed size of n individuals each. Instead of just one trait, the model uses four traits e ; d; e a; α δ; with values constrained to greater than 0 and less than 1, representing an individual’s readiness to invest in the four different categories of behavior in warfare: e ~altruistic attack, d~selfish defense, e a~selfish attack, α δ~altruistic defense. In the following we let fi denote the current fitness of an individual, which at birth is set to 100. Whenever fi drops below 1, the respective individual instantly dies. The aggregate fitness of the individuals of a group will be denoted as FA for the attacking group and FD for defenders. For every hostile encounter of two groups, we can then write that every individual i in the attacking group invests   e i ½1−ai Š f i Ai ¼ Aself ;i þ Aalt;i ¼ ai þ α

Hum Nat

into the attack while every individual j in the defending group invests    δ j 1−d j f j D j ¼ Dself ; j þ Dalt; j ¼ d j þ e into the defensive effort. A completely altruistic attacker would thus, e.g., be charace i = 1, while a completely selfish defender would be characterized terized by ai =0 and α e by dj =1 and δ j = 0. For easier comparison, let us finally define two auxiliary variables   e i ½1−ai Š and δ j ¼ e δ j 1−d j which, analogous to a and d, indicate the relative αi ¼ α amounts of fitness actually invested selfishly and altruistically. It follows directly that 0≤ai +αi ≤1 and 0≤di +δi ≤1 always. The respective efforts of both groups are then given as X X
X A ¼ Aself þ Aalt ¼ Aself ;i þ Aalt;i ¼ ai f i þ ai f i i

and D ¼ Dself þ Dalt ¼

X j

i

i

X
X Dself ; j þ Dalt; j ¼ djf j þ δj f j

j

j

The supremacy s of an attack can then be calculated as s=(A–D)/(A+D), implying -1≤s ≤1, and used to modulate the respective risks and benefits of the hostile encounter. Note that for the case D>A (i.e., when the attack is weaker than the defense), s becomes negative. The model has three features which distinguish altruistic from selfish investments: First, while the individuals on the successful side of a conflict recoup the remainder of their selfish investments that were not needed to overpower the other side (i.e., s Aself,i for successful attackers and |s| Dself,j for successful defenders), altruistic investments are never refunded. Second, altruistic investments do not increase the relative share of the aggregate loot a successful attacker receives. Third, altruistic investments induce twice the risk of dying compared with selfish investments (see “Model Structure: Risks” below). This, of course, is a somewhat arbitrary assumption, but it does fulfill the requirement that there has to be some additional cost to altruism in defense as well. It was made in order not to have to introduce yet another parameter and should be varied in future models. Thus, the payoff π to an attacker i in a successful group of attackers (s>0, Aself >0) is modeled as:
ð1 þ sÞ Aself ;i ⋅L⋅ð F D −DÞ⋅ πi Aself ;i ; Aalt;i ¼ s⋅Aself;i −Aalt;i þ 2 Aself L∈[0,1] is a constant factor determining the minimum percentage of the defenders’ remaining aggregate resources that are stolen in a single raid, namely L/2 if s is close to 0. Accordingly, the maximum percentage is L, when s = 1. Every successful attack thus produces an aggregate benefit of (1+s)⋅L⋅(FD −D)/2, and this aggregate loot is distributed among the attackers proportional to their selfish investments in the attack. For the special case of Aself =0, attackers receive no payoff. Defeated defenders, on the other hand, lose their investments after a hostile encounter and additionally suffer a fair share (1/n) of

Hum Nat

the aggregate loss L FD caused by the raid. This yields the following payoff π to a defender j:
ð1 þ sÞ L⋅ð F D −DÞ ⋅ π j Dself ; j ; Dalt; j ¼ −Dself ; j −Dalt; j − 2 n When an attack fails (s≤0), on the other hand, no benefits are gained by the attackers whereas the defenders recoup the remainder of their selfish investments, and payoffs amount to π i ð⋅Þ ¼ −Aself ;i −Aalt;i for attackers and π j ð⋅Þ ¼ jsj⋅Dself; j −Dalt; j for defenders. Model Structure: Risks In addition to benefits gained and losses suffered, the individuals on all sides are subject to different risks of dying in war. These are modeled as follows. In case of a successful attack (s>0), individuals in the attacking group die with a chance of
Aself;i þ 2⋅Aalt;i ρi Aself;i Aalt;i ¼ ð1−sÞ Aself þ 1 which depends on the superiority of the attack and the investments of the individual relative to the selfish investments of its group. For a nonparticipant i, for example, the risk of dying in the attack is 0, since Aself,i +2 Aalt,i =0 in this case. Note that the constant 1 in the denominators of this and the following expressions is only added to prevent division by zero in the rare cases where Aself,i =0. The reduction of risks introduced by doing so is negligible, since usually 2Aself,i >> (Aself,i +2 Aalt,i). Defeated defenders die with a chance of  
Dself ; j þ 2⋅Dalt; j ρ j Dself ; j ; Dalt; j ¼ ð1 þ sÞ⋅ 1 þ ⋅θ Dself þ 1 which mirrors risks for attackers but has a minimum value defined by a constant θ, 0≤θ≤1, which represents a measure of the expected brutality of the victorious war party. The limiting case of θ=0 represents very “humane” attackers, who only steal but do not murder, while the other extreme, θ=1, implies the complete annihilation of the defeated group. Although crucial to the evolutionary dynamics, only one previous study (Smirnov et al. 2007) explicitly analyzed this factor. Instead it has often even been assumed in modeling that attackers always seek the complete annihilation of the defending group (e.g., Bowles 2009). Risks for the case of a failed attack (s≤0) were modeled in symmetry to the success case, yielding a risk of   Aself ;i þ 2⋅Aalt;i ρi ð⋅Þ ¼ ð1−sÞ⋅ 1 þ ⋅θ Aself þ 1 for unsuccessful attackers and

Hum Nat

ρ j ð ⋅Þ ¼ ð 1 þ s Þ

Dself ; j þ 2⋅Dalt; j Dself þ 1

for successful defenders. The model assumptions made here are quite unfavorable for altruistic investments. Thus, it is not obvious why evolution would select for individuals that invest via α or δ in this model at all. Yet, as will be demonstrated in the following, there are conceivable scenarios in which this can happen, theoretically. Model Structure: Aggression, Risk Aversion, and Deterrence Previous models frequently did not allow groups to pre-assess the strength of their potential opponents and base their decisions on this information. To account for the empirical fact that in human, and also chimpanzee (Watts and Mitani 2001; Wrangham 1999), intergroup conflict attackers seem to quite carefully assess the risks of their undertaking and mostly carry out attacks only when their perceived risk is sufficiently low (Pitman 2011; Wrangham and Glowacki 2012), in the model used here every individual is equipped with a trait γ, 0≤γ≤1, which represents its “aggressiveness” and can evolve independently of the other traits. When two groups meet, the group with the higher average aggressiveness, Γ ¼ 1n ∑i γi , compares its own aggregate investment in attack A with the expected investment in defense of the other group, denoted as Dexp. The more aggressive group then checks if the inequality A−Dexp >(1−Γ) ⋅A holds, meaning that the expected superiority of the attack is greater than the threshold set by (1−Γ) ⋅A. If so, the more aggressive group will attack. For Γ=1 (meaning that every individual of the assessing group is maximally aggressive), that group will attack whenever A−Dexp >0, since (1−Γ) A=0 in that case. If the inequality does not hold, however, the roles of the two groups are reversed and the less aggressive group checks if it should attack using the same calculation. The model also explicitly incorporates deterrence. This is implemented on the group level through a factor ΔX ≥0. This factor is used to modulate the actual defensive strength, DX, of a group X when it is being assessed by another group by defining DX,exp =(1+ΔX) DX. Whenever a group successfully attacks or defends itself, the constant Δ, 0≤Δ≤1, is added to its current deterrence level ΔX. Additionally, to simulate the outdating of deterrence information, the current deterrence level of each group is multiplied by a factor ∂, 0≤∂≤1, each time step during the simulation. The risk assessment mechanism described so far rules out unsuccessful attacks, since Dexp ≥D always. It is more realistic to assume, however, that attackers are not perfect in their ability to assess the risks of an attack, e.g., because they might have insufficient knowledge of the defenders’ territory, yielding a potential advantage even to outnumbered defenders that is difficult to predict (Kelly 2005). To account for this, an error inducing probability ε, 0≤ε≤1, is included in the model. When consecutively assessing the defensive strength of another group, the assessing group unconditionally attacks with probability ε (i.e., irrespective of the possibility that it might be of inferior strength). Finally, if none of the groups attacks, encounters are counted as peaceful. Table 2 gives an overview of the parameters of the model.

Hum Nat Table 2 Overview of central model characteristics Parameter Fixed or Description evolving?

Refer to

a, d

evolving

“bellicosity” and “defensiveness”—i.e., the selfish investments in attacks/defenses

α, δ

evolving

altruistic investments in attacks/defenses

Model Structure: Benefits and Costs

γ

evolving

aggressiveness—i.e., a measure of risk affinity

Model Structure: Aggression, Risk Aversion, Deterrence

L

fixed

minimum impact of raid on fitness of defending group

Model Structure: Benefits and Costs

θ

fixed

brutality of the winning war party

Model Structure: Risks

Δ

fixed

deterrence bonus for successful attackers/defenders

Model Structure: Aggression, Risk Aversion, Deterrence

∂

fixed

outdating rate of deterrence level

Model Structure: Aggression, Risk Aversion, Deterrence

ε

fixed

error probability in risk assessment

Model Structure: Aggression, Risk Aversion, Deterrence

λ

fixed

migration rate

Simulation Cycle

Model Structure: Benefits and Costs

Methods Modeling Technique and Additional Parameters The simulation model was implemented as a Java desktop application. Its source code is available from the author upon request. Except for the newly introduced parameters which were systematically varied during extensive simulation runs, the simulation intentionally uses basically the same parameter values and modeling techniques as one of the most prominent previous modeling studies of ancestral human intergroup conflict, namely the study by Choi and Bowles (2007). This way, the results obtained using the modified model presented here can more easily be compared with previous findings. The same general limitations regarding external validity apply as for all simulation studies. Most important limitations include: (1) Fixed group sizes are assumed for analytical feasibility (g = 20 groups of n = 26 males each). (2) In order to maintain comparability, the same simplified reproduction algorithm is used as in Choi and Bowles (2007): After all wars have been fought in a given time step of the simulation, the dead men in a defeated group are replaced by migrating offspring of males from the group that just defeated them. These males are chosen to reproduce with a probability proportional to their current relative fitness within their group (i.e., including the resources just raided). This mechanism can be interpreted as a kind of group fissioning and fusing: Victorious groups are able send out offspring who then fuse with the remaining members of the defeated groups. Although this is a canonical modeling method, future studies should also analyze models that allow for variable group sizes and actual spatial expansion of territories.

Hum Nat

Simulation Cycle The simulation cycle consists of three main stages. (Stage 1) Outdating and migration: The deterrence levels of all groups are multiplied by ∂ to simulate outdating of deterrence information. Then a fixed percentage, λ, of the population migrates. This is implemented as λ (g n) swaps, 0≤λ≤1, of two individuals randomly chosen from the meta-population. (Stage 2) Group encounters: In the second stage, all groups are randomly paired, yielding 10 random pairings of two groups at each time step. Groups then decide whether to attack or not using the decision rules described above (see “Model Structure: Aggression, Risk Aversion, and Deterrence”). In cases in which no group decides to attack, the encounter is counted as peaceful and the groups do not interact further in this time step. When one group decides to attack, the risks and benefits of the conflict are calculated and distributed as described above (see “Model Structure: Benefits and Costs” and “Model Structure: Risks”). (Stage 3) Baseline mortality and reproduction: After the group interaction stage, a randomly chosen 5% of all individuals and all individuals with an age of 10 time steps in the meta-population die, simulating baseline mortality. Then, the dead men in victorious and peaceful groups are replaced by offspring of their fellow group members, which are chosen to reproduce with a probability proportional to their current relative fitness score. The dead individuals of a defeated group are replaced in the same way, but by offspring of men from the group which just defeated that group. Traits are inherited directly, but small copying errors μ occur, representing random mutations, μ=m x with x~Φ (0,1), m=0.05. Simulated Scenarios The simulation was repeatedly run using systematically varied values of the fixed parameters (see Table 2), representing 48 different scenarios. For each scenario, the simulation was started from 100 different combinations of a and d (i.e., the centers of the 10×10 uniform tiles of the accordingly discretized a×d plane) and then run for 5,000 time steps. This was repeated ten times for every combination of a and d, e; e yielding a total of 5,000,000 observations per scenario. The parameters α δ and γ were always set to 0.0 initially—because this model seeks to discover conditions in which aggressive intergroup conflict and altruism evolve, not conditions in which they are stable once established. Table 3 presents an overview of the scenarios simulated. Full information on all scenarios is available in the ESM. Table 3 Overview of the 48 scenarios simulated Parameter Values used

Loot size

Brutality

Deterrence

Assessment error

Migration

L1 =0.20 L2 =0.60

θ1 =0.15

Δ1 =0.00

ε1 =0.00

λ1 =0.00

θ2 =1.00

(∂1 =0.00)

ε2 =0.25

λ2 =0.20

θ3 =0.35

Δ2 =1.00 (∂2 =0.90)

Count

2

3

2

2

2

Cumulative combinations

2

6

12

24

48

Hum Nat

Simulation Results Finding 1: Observed Evolutionary Outcomes There were three characteristic long-time evolutionary outcomes: (1) a state of peace in which traits drift randomly because the very rare instances of warfare have no noticeable long-term impact on the evolutionary dynamics; (2) a dynamically stable state of “balance of power” (BP) with sporadic attacks in which aggression, bellicosity, and defensiveness are kept at equilibrium by selection; and (3) a state of “immediate surrender” (IS) in which aggression and bellicosity are high while defensiveness remains low. Dynamics with oscillation between two of these states were observed sporadically as well (see, e.g., S47 in the ESM). Figures 1, 2 and 3 show selected simulation scenarios exemplifying the three characteristic states (see the ESM for the complete set of results: S47–S48). The left panels of the figures show the discretized (a×d) space. The darker a tile is shaded, the longer the simulation stayed in this particular (a×d) state. Arrows indicate the expected transitions between (a×d) states. These transition probabilities were estimated from the 5,000,000 observations for each of the 48 scenarios. See Choi and Bowles (2007) for a technical description of how to obtain these estimates. The right panels display the frequencies with which the simulation was observed to be in particular (α×δ) states. The lighter the color of a bar the more peaceful was the population while in this particular (α×δ) state. Most simulations converged to a single state. However, the IS state did appear as a second attractor under a range of conditions. All fixed parameters modeled were found to have significant influences on the evolutionary dynamics. These influences are described qualitatively in the following. Finding 2: Effect of Deterrence The introduction of deterrence has one main effect: it increases the number of peaceful encounters and thus slows down the evolution of bellicosity and defensiveness (i.e., a and d). Especially when combined with higher levels of migration (see immediately

Fig. 1 (Scenario 21; see Finding 1 for description): Random drifting of traits and peace. Parameter values: ε= 0.00, Δ=1.00, ∂=0.90, L=0.20, θ=0.15, λ=0.20

Hum Nat

Fig. 2 (Scenario 4; see Finding 1 for description): Balance of power. Parameter values: ε=0.00, Δ=0.00, ∂=0.00, L=0.60, θ=0.15, λ=0.00

below), deterrence leads to mostly peaceful dynamics. No noticeable systematic effect of deterrence on altruism (i.e., α and δ) was observed. Finding 3: Effects of Migration Like deterrence, migration promotes peace. It does so because it antagonizes stable long-term increases in group-level trait means, reducing the probability for formation of aggressive bellicose groups. Nevertheless, even with a migration frequency of λ=0.20, when assessment errors were introduced, the simulations converged to either BP or IS. Finding 4: Effects of Assessment Error The main effect of assessment error is, of course, that it promotes war. Almost all simulations with assessment error enabled showed high frequencies of conflict. Only

Fig. 3 (Scenario 31; see Finding 1 for description): Immediate surrender. Parameter values: ε=0.25, Δ=1.00, ∂=0.90, L=0.60, θ=0.15, λ=0.20

Hum Nat

when migration, deterrence, and brutality were high but minimum loot size was low at the same time—in other words, when there was not much to gain but very much to lose for aggressive groups—did the population remain peaceful (see S30 in the ESM). Finding 5: The Important Role of Brutality The devastation brought about by victors—whether as a result of their brutality or the lethality of the war technology used—plays a vital role in the simulation. As has been noted (Smirnov et al. 2007), the evolutionary dynamics of intergroup conflict crucially depend on this parameter. The model used here shows that, for lower levels of θ in particular, selection might even favor individuals, and thus groups, who do not defend themselves at all, while still frequently raiding out-groups, as represented by the state of “immediate surrender.” In these scenarios, investing in defense does not pay off individually because the risk of dying upon being defeated is small (down to 0), anyway. On the other hand, when θ is set to high levels and not simultaneously antagonized by sufficiently high levels of migration, warfare can become a “group selective” force. Only in scenarios with θ=1.00 and λ=0.0, i.e., when all defeated defenders were killed by the victors and no migration was possible, did noticeable altruistic investments in attacks evolve at all in this simulation (S10 and S12; also see Finding 6). Finding 6: Altruism Might Be More Likely To Evolve for Defense In those simulations where conflict did occur and traits did not just drift randomly, altruism in defense, δ, was more likely to evolve than altruism in attacks, α. As discussed above (see “An Alternative Payoff Structure”), the model reflects the intuition that in defense, altruistic investments are not selected against as strongly. Since investments here are likely to secure the individual’s own survival and that of its kin, overshooting the goal of successful defense a little by taking greater risks than absolutely necessary is not as strongly selected against as long as wars are not too frequent (or attackers “too humane,” which would make surrender the fitnessmaximizing strategy for defenders). Even in the extreme case of θ=1, only a very slight difference exists between selfish and altruistic investments in defense because failed defenses definitely result in an individual’s death. Consequently, δ reaches fairly high levels in these scenarios (see, e.g., S12 and S16). Selfish investments in defense, however, evolve to higher levels than altruistic ones in all scenarios since they entail only half the chance of dying in case of a successful defense. Altruism in attacks, on the other hand, could only persist when an extreme level of brutality was assumed, no migration took place, and war was frequent. But this result has to be interpreted with caution: in scenarios without migration and extreme levels of devastation for defeated groups—which, in my view, classify as “old” group selection scenarios in West et al.’s (2011) terminology—it is crucial for individual survival to be part of the more aggressive and stronger party in a conflict because then, literally, attack is the best form of defense. This reduces the difference in risks between selfish and altruistic investments in attack to some degree because the very likely alternative to both becomes death, thus transforming attack into a threshold defense PGG as well.

Hum Nat

Discussion To summarize central modeling results: With risk aversion, migration, and deterrence in place, intergroup tolerance rather than merely conflict is frequently observed. For a broad range of parameter values, altruism in attacks is strongly selected against, as was expected from the outset. Only the empirically rather unmotivated but theoretically possible reintroduction of “old” group selection (high brutality, low migration, and frequent wars; see West et al. 2011) allows for its evolution in this model, but even then it reaches only low levels in terms of relative fitness invested, thus still being far from self-sacrifice. Altruism in defense, on the other hand, evolved to moderate levels in some of the more realistic scenarios. Thus, perhaps most importantly, the model shows that depending on the choice of the exogenous parameters a multitude of evolutionary dynamics is possible under the assumptions made here. A number of these assumptions, however, need to be critically considered before drawing conclusions from the observations made. First, the model presented here, just like earlier models, still uses many simplifications. These include fixed group sizes, random migration and random group matching, and no explicitly modeled kinship structure. In particular, it is doubtful, for example, that the brutality of attackers and defenders has exactly the same degree, as is assumed here. Successful defenders might not be able to carry out as much violence because defeated attackers will quickly retreat to their own territory. Changing this assumption, however, would likely only yield higher expected returns for attackers and thus not change dynamics substantially. Furthermore, the assumption of completely random migration is unrealistic. It seems more likely that individuals will selectively migrate to prosperous groups and stay longer with them. Second, a more important deficit of the model is that it does not incorporate competition for status and mates within groups. The model’s purpose is mainly to investigate the changes in the evolutionary dynamics of the interplay of bellicosity and altruism which result from refining earlier PGG models. Future models should focus more on the influences of within-group status competition and sexual selection on bellicosity (van Vugt et al. 2007). Finally, one additional feature of the current model is worthy of discussion (see “Is War an n-Person Prisoner’s Dilemma?”). A crucial change from earlier models consists of the alternative distribution rule for the benefits produced by attacks in the current model. In congruence with empirical observations (Chagnon 1988), the individual payoff to attackers here is performance-linked. This contrasts with earlier PGG models, which assume a fixed marginal per capita return for all group members. Thus, eventually, both modeling approaches introduce a very crucial characteristic simply by assumption. It would be a very promising enterprise to investigate how distribution rules of war benefits evolve endogenously in heterogeneous populations.

Outlook The arguments and findings presented here should help to curtail exuberant interpretations of previous models that are based on an obviously oversimplified PGG view of the nature of ancestral warfare and sometimes assume extreme levels of group

Hum Nat

selection. Models such as the one presented here, which try to get a bit closer to a realistic reconstruction of the risks and benefits of ancestral warfare, may be better qualified to analyze the complex patterns of prehistoric human intergroup conflict and intergroup tolerance (Boehm 2012; Durham 1976; Fry and Söderberg 2013; Wrangham and Glowacki 2012). The frequently resulting state of “balance of power” with sporadic attacks, e.g., might capture the basic logic of real hunter-gatherer warfare quite well. Nonetheless, the model at hand cannot, of course, offer more than one step in this direction. However, at least two empirical follow-up questions are highlighted by this model. (1) Does the state of “immediate surrender,” which could also be called “tolerated intergroup theft,” correspond to real patterns of intergroup conflict? At least one aspect of real conflict comes to mind which might fit this prediction: ritualized appeasement through gifts—for example, Danegeld (Boehm 2012). In some scenarios it simply might not pay for defenders to risk engaging in violent conflict. They might rather accept a certain loss of resources which, compared with fighting, has smaller fitness consequences. (2) What are the factors determining the strength of the brutality, θ, in real conflicts? That is, did our ancestors really engage in such murderous frenzies as represented by high values of θ? And if so, when did they and when did they not? The available studies on chimpanzee and hunter-gatherer warfare are not univocal about this point. In some reported instances chimpanzee warfare led to the complete annihilation of the losing group (Wrangham 1999), in others it did not. In some cases of hunter-gatherer intergroup conflict, the conflict is so severe that both groups lose members (Beckerman et al. 2009); in other circumstances, peace is the rule (Boehm 2012) or only minor skirmishes take place (Fry and Söderberg 2013; also see Bowles 2009; Durham 1976; Walker and Bailey 2013). Some factors likely to be important for determining the value of θ, and also the importance of deterrence (i.e., Δ and ∂), are resource scarceness and geographical boundaries limiting territory size and group movement. These factors, of course, vary between habitats, which might explain some of the variance in the levels of intergroup aggression and conflict found in the anthropological record (Keeley 1996), but also in currently violent regions (see, e.g., Gebre-Michael et al. 2005). In order to understand our evolved motivational psychology for intergroup aggression better—and maybe even to find better ways to prevent circumstances leading to outbreaks in comparable groups today—answering these questions is of great importance. Finally, if warfare should turn out not to be a strong driver in the evolution of cooperation and altruism, because, as has been argued here, the link between altruism and intergroup aggression might be weaker than assumed, the coevolution hypothesis of altruism and bellicosity could be called into question. We would then need alternative—and independent—explanations for the evolution of bellicosity and in-group altruism. Aggressive bellicosity, as has been shown above, can be (very) beneficial individually if it is accompanied by careful risk assessments, which very likely is the case with chimpanzees (Watts and Mitani 2001; Wrangham 1999; Wrangham and Glowacki 2012). Group defense, on the other hand, while also entailing direct benefits to the individual, namely survival and resource protection, likely offers more opportunities for natural selection to favor altruistic behavior (see Rusch 2013).

Hum Nat

Another strong hint on the potential evolutionary history of intergroup conflict is its “sexual dimorphism.” In chimpanzees, just as in humans, warfare is almost exclusively a male occupation. Chagnon’s classical finding that Yanomami warriors who had demonstrated their killing abilities in warfare have increased reproductive success (Chagnon 1988) would support this view. However, more recent observations from other hunter-gatherer groups show that things might not be as simple (Beckerman et al. 2009). Unfortunately, not much data have been published on one central variable which could shed more light on this question: In order for bellicosity to spread in a population, it might not be necessary to assume that warriors actually have more offspring; if wars are frequent enough, and the risk of dying in defense is shared equally among the members of the group under attack, it might suffice if warriors reproduce earlier than the others, and thus “outreproduce” their non-bellicose fellow group members. From the study of demographic data on recent wars (WWII and Vietnam) it is known, at least, that veterans marry significantly sooner after returning home than non-veterans of the same age (Modell and Steffey 1988; Modell and Haggerty 1991). Whether or not this finding from modern warfare can be explained by resorting to our evolved psychology for intergroup aggression in small group contexts is an open question, though. Alternative explanations of the evolution of human general cooperativeness and altruism have been proposed (see, e.g., Tomasello et al. 2012; West et al. 2007), but this is not the place to argue for one over the others. An important guide in the search for explanations, however, might be the observation that there are no significant indications that general cooperativeness differs strongly between the sexes (Balliet et al. 2011). The adaptive problems that initiated the evolution of cooperativeness will therefore likely have been problems with which both sexes were confronted, e.g., successfully reaping the benefits of all forms of division of labor, including cooperative breeding (Jaeggi et al. 2010). The evolution of such cooperative abilities was likely propelled not least by the increase in human intelligence (Brosnan et al. 2010), and they later became almost indispensable because of the specialization made possible by human cultural abilities (Henrich and Boyd 2008). Acknowledgments I thank Max Albert, Charlotte Störmer, Eckart Voland, and Human Nature’s anonymous reviewers for very valuable criticism. Prudent copy-editing by Mary June-el Piper is gratefully acknowledged. This publication represents a component of my doctoral thesis (Dr. rer. nat.) in the Faculty of Biology at the Justus-Liebig-University Giessen, Germany.

References Bach, L. A., Helvik, T., & Christiansen, F. B. (2006). The evolution of n-player cooperation—threshold games and ESS bifurcations. Journal of Theoretical Biology, 238, 426–434. Balliet, D., Li, N. P., Macfarlan, S. J., & van Vugt, M. (2011). Sex differences in cooperation: a meta-analytic review of social dilemmas. Psychological Bulletin, 137, 881–909. Beckerman, S., Erickson, P. I., Yost, J., Regalado, J., Jaramillo, L., Sparks, C., et al. (2009). Life histories, blood revenge, and reproductive success among the Waorani of Ecuador. Proceedings of the National Academy of Sciences, 106, 8134–8139. Boehm, C. (2012). Ancestral hierarchy and conflict. Science, 336, 844–847.

Hum Nat Bowles, S. (2009). Did warfare among ancestral hunter-gatherers affect the evolution of human social behaviors? Science, 324, 1293–1298. Brosnan, S. F., Salwiczek, L., & Bshary, R. (2010). The interplay of cognition and cooperation. Philosophical Transactions of the Royal Society, B: Biological Sciences, 365, 2699–2710. Chagnon, N. A. (1988). Life histories, blood revenge, and warfare in a tribal population. Science, 239, 985– 992. Choi, J.-K., & Bowles, S. (2007). The coevolution of parochial altruism and war. Science, 318, 636–640. Durham, W. H. (1976). Resource competition and human aggression: Part I: a review of primitive war. Quarterly Review of Biology, 51, 385–415. Fry, D. P., & Söderberg, P. (2013). Lethal aggression in mobile forager bands and implications for the origins of war. Science, 341, 270–273. Gat, A. (2009). So why do people fight? Evolutionary theory and the causes of war. European Journal of International Relations, 15, 571–599. Gebre-Michael, Y., Hadgu, K., & Ambaye, Z. (2005). Addressing pastoralist conflict in Ethiopia: The case of the Kuraz and Hamer sub-districts of South Omo zone. Saferworld. http://www.saferworld.org.uk/ resources/view-resource/106. Accessed 9 July 2013. Ginges, J., & Atran, S. (2011). War as a moral imperative (not just practical politics by other means). Proceedings of the Royal Society B – Biological Sciences, 278, 2930–2938. Glowacki, L., & Wrangham, R. W. (2013). The role of rewards in motivating participation in simple warfare. Human Nature, 24, 444–460. Gneezy, A., & Fessler, D. M. T. (2011). Conflict, sticks and carrots: war increases prosocial punishments and rewards. Proceedings of the Royal Society B: Biological Sciences, 279, 219–223. Goodall, J. (1986). The chimpanzees of Gombe: Patterns of behavior. Cambridge: Belknap Press of Harvard University Press. Halevy, N., Weisel, O., & Bornstein, G. (2012). “In-Group Love” and “Out-Group Hate” in repeated interaction between groups. Journal of Behavioral Decision Making, 25, 188–195. Henrich, J., & Boyd, R. (2008). Division of labor, economic specialization, and the evolution of social stratification. Current Anthropology, 49, 715–724. doi:10.1086/587889. Hill, K. R., Walker, R. S., Bozicevic, M., Eder, J., Headland, T., Hewlett, B. S., et al. (2011). Co-residence patterns in hunter-gatherer societies show unique human social structure. Science, 331, 1286–1289. Jaeggi, A. V., Burkart, J. M., & van Schaik, C. P. (2010). On the psychology of cooperation in humans and other primates: combining the natural history and experimental evidence of prosociality. Philosophical Transactions of the Royal Society B, 365, 2723–2735. Keeley, L. H. (1996). War before civilization: The myth of the peaceful savage. New York: Oxford University Press. Kelly, R. C. (2005). The evolution of lethal intergroup violence. Proceedings of the National Academy of Sciences, 102, 15294–15298. Kerr, B., & Godfrey-Smith, P. (2002). Individualist and multi-level perspectives on selection in structured populations. Biology and Philosophy, 17, 477–517. Kohler, T. A., & Kramer Turner, K. (2006). Raiding for women in the pre‐hispanic Northern Pueblo Southwest? A pilot examination. Current Anthropology, 47, 1035–1045. Konrad, K. A., & Morath, F. (2012). Evolutionarily stable in-group favoritism and out-group spite in intergroup conflict. Journal of Theoretical Biology, 306, 61–67. Lehmann, L., & Feldman, M. W. (2008). War and the evolution of belligerence and bravery. Proceedings of the Royal Society B: Biological Sciences, 275, 2877–2885. Lehmann, L., Keller, L., West, S. A., & Roze, D. (2007). Group selection and kin selection: two concepts but one process. Proceedings of the National Academy of Sciences, 104, 6736–6739. Mason, W. A., & Mendoza, S. P. (1993). Primate social conflict. Albany: State University of New York Press. Mathew, S., & Boyd, R. (2011). Punishment sustains large-scale cooperation in prestate warfare. Proceedings of the National Academy of Sciences (USA), 108, 11375–11380. McDonald, M. M., Navarrete, C. D., & van Vugt, M. (2012). Evolution and the psychology of intergroup conflict: the male warrior hypothesis. Philosophical Transactions of the Royal Society, B: Biological Sciences, 367, 670–679. Modell, J., & Haggerty, T. (1991). The social impact of war. Annual Review of Sociology, 17, 205–224. Modell, J., & Steffey, D. (1988). Waging war and marriage: military service and family formation, 1940–1950. Journal of Family History, 13, 195–218. Pitman, G. R. (2011). The evolution of human warfare. Philosophy of the Social Sciences, 41, 352–379. Pratto, F., Sidanius, J., & Levin, S. (2006). Social dominance theory and the dynamics of intergroup relations: taking stock and looking forward. European Review of Social Psychology, 17, 271–320.

Hum Nat Rusch, H. (2013). Asymmetries in altruistic behavior during violent intergroup conflict. Evolutionary Psychology, 11, 973–993. Rusch, H. (2014). A threshold for biological altruism in public goods games played in groups including kin. MAGKS Joint Discussion Paper Series in Economics, No. 29-2014. Saaksvuori, L., Mappes, T., & Puurtinen, M. (2011). Costly punishment prevails in intergroup conflict. Proceedings of the Royal Society B: Biological Sciences, 278, 3428–3436. Smirnov, O., Arrow, H., Kennett, D., & Orbell, J. (2007). Ancestral war and the evolutionary origins of “Heroism.” Journal of Politics, 69, 927–940. Tomasello, M., Melis, A. P., Tennie, C., Wyman, E., & Herrmann, E. (2012). Two key steps in the evolution of human cooperation. Current Anthropology, 53, 673–692. Tooby, J., & Cosmides, L. (2010). Groups in mind: The coalitional roots of war and morality. In H. HøghOlesen (Ed.), Human morality and sociality: Evolutionary and comparative perspectives (pp. 191–234). Basingstoke: Palgrave Macmillan. van Vugt, M., de Cremer, D., & Janssen, D. P. (2007). Gender differences in cooperation and competition: the male-warrior hypothesis. Psychological Science, 18, 19–23. Walker, R. S., & Bailey, D. H. (2013). Body counts in lowland South American violence. Evolution and Human Behavior, 34, 29–34. Watts, D., & Mitani, J. (2001). Boundary patrols and intergroup encounters in wild chimpanzees. Behaviour, 138, 299–327. West, S. A., Griffin, A. S., & Gardner, A. (2007). Social semantics: altruism, cooperation, mutualism, strong reciprocity and group selection. Journal of Evolutionary Biology, 20, 415–432. West, S. A., El Mouden, C., & Gardner, A. (2011). Sixteen common misconceptions about the evolution of cooperation in humans. Evolution and Human Behavior, 32, 231–262. Wrangham, R. W. (1999). Evolution of coalitionary killing. American Journal of Physical Anthropology, 110, 1–30. Wrangham, R. W., & Glowacki, L. (2012). Intergroup aggression in chimpanzees and war in nomadic huntergatherers. Human Nature, 23, 5–29. Hannes Rusch studied philosophy and mathematics. He works in the fields of behavioral economics, philosophy of biology and science, evolutionary psychology, and business ethics. In addition to his position for the Chair of Behavioral and Institutional Economics he works for the Chair of Philosophy of Biology at JLU Giessen and the Peter Löscher Chair of Business Ethics at TU München.