Love or fear: Can punishment promote cooperation?

Cooperation is a paradox: Why should one perform a costly behavior only to increase the fitness of another? Human societies, in which individuals cooperate with genetically unrelated individuals on a considerably larger scale than most mammals do, are especially puzzling in this regard. Recently, the threat of punishment has been given substantial attention as one of the mechanisms that could help sustain human cooperation in such situations. Nevertheless, using punishment to explain cooperation only leads to further questions: Why spend precious resources to penalize free‐riders, especially if others can avoid this investment and cheaters can punish you back? Here, it is argued that current evidence supports punishment as an efficient means for the maintenance of cooperation, and that the gravity of proposed limitations of punishment for maintaining cooperation may have been overestimated in previous studies due to the features of experimental design. Most notably, the importance of factors as characteristic of human societies as reputation and language has been greatly neglected. Ironically, it was largely the combination of the two that enabled humans to shape costly punishment into numerous low‐cost and less detrimental strategies that clearly can promote human cooperation.


SEBESTIAN KROUPA
Cooperation is a paradox: Why should one perform a costly behavior only to increase the fitness of another? Human societies, in which individuals cooperate with genetically unrelated individuals on a considerably larger scale than most mammals do, are especially puzzling in this regard. Recently, the threat of punishment has been given substantial attention as one of the mechanisms that could help sustain human cooperation in such situations. Nevertheless, using punishment to explain cooperation only leads to further questions: Why spend precious resources to penalize free-riders, especially if others can avoid this investment and cheaters can punish you back? Here, it is argued that current evidence supports punishment as an efficient means for the maintenance of cooperation, and that the gravity of proposed limitations of punishment for maintaining cooperation may have been overestimated in previous studies due to the features of experimental design. Most notably, the importance of factors as characteristic of human societies as reputation and language has been greatly neglected. Ironically, it was largely the combination of the two that enabled humans to shape costly punishment into numerous low-cost and less detrimental strategies that clearly can promote human cooperation.
"Upon this a question arises: whether it be better to be loved than feared or feared than loved? It may be answered that one should wish to be both, but, because it is difficult to unite them in one person, [. . .] either must be dispensed with.
[. . .] Love is preserved by the link of obligation which, owing to the baseness of men, is broken at every opportunity for their advantage; but fear preserves you by a dread of punishment which never fails." Niccol o Machiavelli, The Prince, ch. XVII. p 57-58. In his last presidential address to the Royal Society in 2005, Robert May claimed that "the most important unanswered question in evolutionary biology, and more generally in the social sciences, is how cooperative behaviour evolved and can be maintained." 1 Humans are particularly puzzling in this regard, as they are known to cooperate with genetically unrelated individuals on a substantially larger scale than do most mammal species. [2][3][4] Recently, the notion of costly punishment has been hotly discussed as one of the factors that could boost human cooperation. 2,[5][6][7][8] Following Clutton-Brock and Parker (1995), punishment is a "retaliatory infliction of fitness reduction . . . which benefits the punisher because it reduces the probability that the victim will repeat a damaging action or will refuse to perform a beneficial one." 9:209 This definition is now widely accepted. 6,10,11 Punishment is therefore a temporarily spiteful strategy, which nonetheless yields delayed benefits because the recipient's behavior becomes more cooperative in subsequent interactions (Fig. 1).
To illustrate this strategy, one of the best pieces of evidence of punishment in nonhuman species comes from the mutualistic interaction between the blue-streak cleaner wrasse (Labroides dimidiatus) and its fish clients. Cleaners are expected to remove ectoparasites and dead tissue, but occasionally cheat by feeding on mucous on their client's skin, thus harming the client. In a series of experiments, it has been observed that if clients have been cheated, they often aggressively chase the cleaner. 12 Although it is instantly costly to the client, this behavior brings clear future benefits, since cleaners thus punished are less likely to cheat in subsequent interactions.

PUNISHMENT AMONG HUMANS
Some of the strongest evidence that humans are motivated to punish cheaters and that punishment promotes cooperation comes from diverse economic games. First, in the ultimatum game, a sum of money is allotted to two anonymous players and one of them proposes how to split the stake between the two. The Sebestian Kroupa recently graduated with a degree in biology from the University of St. Andrews and has since enrolled in a graduate program in the Department of History and Philosophy of Science at the University of Cambridge. His research interests revolve around the history of evolutionary thought. E-mail: sk796@cam.ac.uk second player can either accept the offer, in which case the players keep the money, or reject it, in which case both players receive nothing. To avoid reciprocation, the game is played only once. The second player, if acting according to economic rationality, should accept any nonzero offer, as the chimpanzee (Pan troglodytes) typically does in this game. 13 When the game is played by humans, however, most offers below 20%-30% are rejected. 14,15 This trend has been observed across a range of cultures from all six continents and thus seems to be a universal human trait (Fig. 2). In light of this evidence, one could argue that people are willing to punish unfair behavior even if the penalty is costly in terms of resources.
The third-party punishment game presents a similar scenario. Three anonymous players receive a sum of money, half of which is allocated to the third player. Player one then proposes how to split the remaining half between the first two players. While the second player is relegated to the role of a mere spectator, the third player can invest a part of his stake to decrease player one's payoff, should he consider his offer to player two unfair. Although a self-interested or neutral player three should never pay to penalize player one, crosscultural studies indicate that, as in the ultimatum game, offers below 20%-30% are commonly punished (Fig. 3). 15,16 Consequently, even as uninterested third parties, humans appear to be willing to inflict costly punishment.
While results from the ultimatum and third-party punishment games indicate that people are motivated to punish even if such penalties are costly, evidence from the public goods game demonstrates that the punishment of selfish individuals promotes and stabilizes cooperation at higher levels than if no penalties are available. 2,5,17,18 In the public goods game, players can either cooperate by contributing to the common pool, which is multiplied and evenly redistributed among all members at each turn, or defect, thereby refusing to share their endowment. Selfinterested individuals typically realize that by withholding their tokens, they can achieve higher payoffs and that, in consequence, average contributions quickly decrease close to zero. Nevertheless, if costly punishment of other players is allowed -that is, if individuals can pay a cost to reduce the payoff to others -free-riders are, heavily penalized by the group and average contributions approach the maximum value (Fig. 4).
Further findings highlight the strength of the motivation to punish, since humans impose costly penalties on cheaters regardless of whether they engage in repeated or one-shot anonymous encounters, although the latter promise no possibility of subsequent cooperative interactions. 18 Moreover, humans are eager to pay even for only symbolic punishment, or rather, disapproval, which does not directly harm the free-riding individual. 19 On the other hand, penalties are, to some degree, still subject to economic rationality, since the willingness to inflict punishment decreases as its cost grows. 8,20,21 It has been documented that the degree of costly punishment positively co-varies with the extent of cooperation across cultures, 15 supporting the notion that penalties are important in the maintenance of cooperation. It might seem that the positive effect of punishment on sustaining cooperation can be hardly doubted.
The Hazardous Nature of Costly Punishment "In the state of nature there often wants power to back and support the sentence when right, . . . resistance many times makes the punishment dangerous, and frequently destructive, to those who attempt it." John Locke, Second Treatise of Civil Government 91:66 .
Although ample evidence supports the role of punishment in the maintenance of cooperation, this strategy . Rejection rates in the ultimatum game across a set of distinct cultures. The area of the black bubbles displays the fraction of player 2s who rejected a given offer. The numbers inside indicate the percentage illustrated by that bubble. Replicated, with permission, from Henrich and colleagues. 15 has been widely criticized as too costly and destructive. [21][22][23] Punishment is effective in promoting cooperation, since the threat of harm makes virtually everyone contribute highly.
However, cooperation enforced by punishment is largely inefficient because the costs of punishment may exceed the benefits of cooperation. 21,[23][24][25] Once costly punishment is extensively used, average group payoffs may decrease considerably, since both the punishing and the punished lose resources. In consequence, net payoffs in groups with punishment are often lower than payoffs in groups without the option to punish (Fig. 5).
Moreover, punishers are subject to exploitation by second-order freeriders who do cooperate, but fail to pay to penalize cheaters and therefore achieve higher payoffs than those who punish. [26][27][28] In this respect, consider the following question: Why sacrifice for the public good and punish, especially if the act is costly? Obviously, some individuals, even if they are otherwise cooperative, might be tempted to avoid this investment.
Furthermore, although initial models and studies investigating the effects of costly punishment on cooperation had typically assumed that penalties are targeted exclusively at cheaters, this need not be the case. Studies have now shown that punishment can be directed not only at free-riders, but also at cooperators or punishers. This so-called antisocial punishment interferes with the beneficial influence of the punishment of cheaters, hindering both the effectiveness and efficiency of cooperation, which breaks down and yields even lower benefits. 25,[29][30][31][32] As punishers of cheaters themselves become the target of punishment, they obviously are less likely to punish free-riders in return.
This phenomenon is especially prevalent when an additional round of punishment is added to the game by allowing counter-punishment. 29,30,32 Correspondingly, antisocial punish-ment is thought to be motivated primarily by revenge, as punished cheaters feel the urge to get back at their punishers. 18,25,30 Alternatively, this behavior can be driven by a dislike of "do-gooders" or feelings of humiliation in the context of more generous contributions. 18 Vengeful antisocial punishment can be equally motivated by strategic considerations, since punishers of cheaters are less likely to inflict penalties in the future if they are punished for that behavior in the first place. 30 In this manner, by signaling that future penalties will be met with counter-punishment, antisocial punishers can afford to contribute less and thus free-ride in subsequent rounds. To add an underlying evolutionary explanation to these proximate motives, since antisocial punishment can ultimately improve the relative payoff of an actor, it can be favored by selection. 31 Experimental evidence clearly shows that antisocial punishment presents a real threat to cooperation. Typically, 20% or more of all punitive acts are directed at high contributors, 29,33 while about 25% of punishments targeted at cheaters are retaliated. 30 Antisocial punishment is common across a wide range of cultures, correlating negatively with the strength of rule of law and civic norms of cooperation. 25 Therefore, using costly punishment to explain the motivation behind cooperation leads to further questions: Why spend precious resources to penalize others, especially if second-order free-riders can avoid this investment  The crucial issue of present research entails its relevance to reallife situations. Because of a mismatch between experimental conditions and the situations in which humans typically use punishment in everyday circumstances, the efficiency of punishment could be inadvertently underestimated. In many cases, evidence that punishment undermines the efficiency of cooperation has come from studies in which the public goods game ran for ten to twenty rounds. 21,22,25,30,34 However, if the game is allowed to run over a higher number of periods, although average earnings are initially higher in the penalty-free condition, punishment becomes less detrimental over time, eventually yielding cooperation that is both highly effective and efficient (Fig.  6). 8,18,35,36 This occurs as a result of a steep decline in the frequency of punishment over successive rounds (Fig. 7). Punishment still remains available as an option when it is necessary to discipline occasional defectors, but there is no longer any need for its extensive use. Since the resources available are not depleted on extensive penalties and the degree of contributions constantly remains high, overall payoffs increase. However, even if rarely applied, punishment still holds the key for cooperation: as soon as it is removed, cooperation deteriorates. 2,5 In the long run, the beneficial influence of punishment on the effectiveness of cooperation persists, but its negative effect on efficiency wanes away.
Why does the frequency of punishment drop over time? The likely explanation is rather simple: People gradually learn the rules of the game or, in other words, punishment requires several rounds to modify the behavior of its recipients, establish as a shared expectation, and reach equilibrium. 8,37 Once this has occurred, participants typically have observed or directly learned that free-riding will not be tolerated by the group. They also have become aware that there is no benefit in cheating. The actual enactment of punishment thus gradually declines as the threat of its infliction is established. Experimental evidence shows that repeated interactions are likely important in this process, allowing the formation of expected future behavior, as well as credibility between players. 8 The initial rounds of the game could thus be regarded as artificial, since the collective strategy has not yet been agreed on or enforced through punishment. This also means that if punishment is truly effective in sustaining cooperation, it should quickly reach equilibrium, where penalties are only seldom used. Therefore, the absence of extensive, costly punishment in a given population could suggest an especially strong local effect; that is, that most cheaters have already modified their behavior and do not need to be penalized again. This could be the case of real-world populations where, since common Figure 5. The efficiency of cooperation in the public goods game when punishment is available (sanction) or absent (baseline). Replicated, with permission, from Sefton and colleagues. 34 Figure 6. The efficiency of cooperation in a public goods game when punishment is available (P) or absent (N). The games were run for either ten or fifty rounds. The number in the parentheses indicates average net earnings in each condition. Replicated, with permission, from G€ achter and colleagues. 35 expectations have typically been established and cheaters know what to expect, punishment should be carried out rarely and thus bear minimal overall costs. The availability of the option to punish, however, is essential if this state is to be maintained. To illustrate this point, although the vast majority of taxpayers are never fined for tax evasion, the existence of this measure clearly discourages people from cheating. As experimental studies demonstrate, if the threat of punishment is diminished or removed, tax evasion increases. [38][39][40] No Choice But to Punish Experiments investigating punishment commonly take the form of a public goods game in which one can pay to impose costs on cheaters. However, in the absence of an alternative strategy, the frequency of costly punishment could be accidentally overrepresented, since participants have no other option to discipline cheaters except through costly punishment. 41,42 In other words, since no intermediate form of punishment is available, players can choose only between two radical options: use a highly costly strategy that might induce overly strong damage or abstain from using it. While the former destroys a large portion of the resources available and negatively affects the efficiency of cooper-ation, the latter cannot sustain cooperation at all.
This point bears special relevance to modeling studies, which frequently use evidently inaccurate payoff matrices. In this respect, it is essential to realize that punishment satisfies its objective when it reduces the cheater's payoff to just below the population average: at this precise point, cooperation becomes more profitable than free-riding. 43 Therefore, any penalties stronger than this are unnecessary and merely damage the public good without soliciting higher contributions, while any investments lower than this yield no benefits at all. Such misrepresentation of the relationship between the costs and the benefits of punishment could lead to an underestimation of its efficiency.
Similarly, in real-world conditions, punishment is not limited to the two extremes and intermediate forms are available. Moreover, situations outside of the lab offer additional alternatives that could help control the hazardous nature of punishment, such as communication and freedom of choice, whereby participants can agree on their decisions, expectations and punitive rules. Accordingly, as the current body of evidence suggests, higher payoffs are achieved if punishment is combined with communication than if the opportunity for dialogue is absent. 44,45 Studies allowing endogenous group formation and rule selection, where players can decide whether to join a group in which costly punishment is allowed or banished, and discuss how and against whom to use penalties, illustrate how costly punishment is handled if players are not as constrained by the experimental design but, instead, given the freedom to choose between strategies and partners. 33,44,46 This direction of research solved the essential question of whether people would deliberately adopt costly punishment.
The answer is affirmative, but with a curious twist. Initially, participants are usually reluctant to allow costly punishment of others or join a group with punitive rules. However, the experience of free-riding gradually convinces them that such measures are necessary if cooperation is to be sustained. 33,44,47 At the end of the game, virtually all groups would adopt costly punishment or all players would migrate to a group in which the option of punishment is available. Moreover, these groups would achieve the highest payoffs and lowest levels of defection. 47 No group ever voted to authorize penalties on high contributors, thus instantly banning antisocial punishment. 33 When participants were allowed voluntarily to associate with players of their preference, and thus determine their own group composition every few turns, cooperators gathered together and free-riders were forced to act as cooperators if they wished to be involved at all. 46 Nonetheless, the option of penalizing free-riders was equally vital, as only the combination of regrouping and costly punishment yielded the strongest cooperation. Interestingly, merely the effect of choice could boost cooperation. In spite of there being essentially no difference between endogenously and exogenously implemented rules, it has been shown that the former yields higher payoffs. 48 The key advantage of the design discussed could be that since participants can openly agree on their shared strategy or join a group where rules have been firmly established, the aforementioned initial artificial phase of the experiment, in which no collective strategy exists, is skipped or considerably reduced and equilibrium is reached instantaneously. In other words, participants do not need to form common expectations of cooperation and punishment through trial and error, as in the classic form of a public goods game, but instead arrive at them through the faster and cheaper means of communication and deliberation. The results of studies allowing for endogenous decision making could be thus interpreted as supporting the previously discussed notion of equilibrium. Considering that communication is normally available in real-life situations, it would be desirable to take this factor into account when investigating the effect of punishment on cooperation among humans.
How Relevant Is Second-Order Free-Riding?
As stated, nonpunishing cooperators avoid the cost of punishment and thus achieve higher average payoffs than do punishing cooperators. Therefore, the question is why sacrifice for the general good and punish. Obviously, some individuals, termed second-order free-riders, might be tempted to avoid this investment and thus achieve higher payoffs. The essence of this problem is based on the assumption that the benefits of punishment are shared equally among all members of a given population. 49,50 However, this is most likely incorrect. In terms of reputation, 26,51 for example, a disproportionate share of the benefits could erase or surpass the difference in payoffs between punishing and nonpunishing cooperators and thus solve the issue.
Although precise evidence for such additional benefits is absent, current studies demonstrate only negligible occurrence of second-order free-riding in humans. This behavior provoked very little anger among participants who were largely unwilling to penalize such behavior. 32,52 Moreover, an observational field study in the Ju/ 'hoansi Bushmen showed that secondorder free-riding never became a subject of conflict. 53 Briefly, the issue appears to bear little or no relevance to real-world circumstances.
The question is, therefore, how humans evolved to solve this problem. On one hand, its significance may have been exaggerated by the use of supposedly inappropriate payoff matrices that assume equal distribution of the benefits of punishment and therefore disregard, for example, the effect of reputation. However, coordination and institutions could also hold the key, since the whole community may be expected to participate and share the costs or, perhaps, appoint and eventually grant professional status to specific punishers.
When interpreting evidence from studies investigating the effects of antisocial punishment, it is impossible not to consider the role of anonymity. The vast majority of experiments on punishment are conducted with subjects under conditions of strict anonymity. Players typically communicate by computer and are informed of the contributions of others penalties to them onscreen, with their identities subsequently reshuffled. 2,5 Therefore, it is truly impossible for players to establish any permanent links with other players and their behavior, which prevents any influence on reputation.
The impact of anonymity on the extent of antisocial punishment remains unclear and controversial. Critics of costly punishment argue that anonymity protects punishers from receiving additional vengeful penalties from the cheaters they had previously punished, so that the rules of the experimental design conceal the full effect of antisocial punishment. 23,30,41 However, anonymity might well contribute to exaggeration of the incidence of antisocial punishment. It has been long known that the perception of anonymity combined with feelings of reduced individual responsibility can produce high rates of misconduct, including aggressive or punitive actions. 54,55 As indicated, experiments investigating antisocial punishment typically take place in analogous conditions. Further, the use of technology can encourage antisocial behavior that individuals otherwise probably would not engage in, such as online trolling. 56,57 On that note, perceptual differences when interacting with a human or a computer partner in economic games have been reported, with the former provoking a stronger neurocognitive response. 58 Consequently, it seems that anonymity could protect antisocial punishers rather than punishers of cheaters.
To resolve this debate, it is essential to do further experiments, ideally in conditions where participants cannot hide behind anonymity. Although studies have been carried out under such circumstances, 46,59 none have looked directly at the interaction between lack of anonymity and antisocial punishment. Yet, the introduction of identity, reputation, and personal contact seems unlikely to favor antisocial behavior.

The Power of Forgiveness
Retaliatory aggression is frequent among most social animals, 9 so why, in light of Mahatma Gandhi's famous quote, is not the whole world blind? In this regard, reconciliation, a common and simple strategy that could bear particular relevance to vengeful antisocial punishment among humans, has been overlooked. This behavior, which has been documented among social primates and social canines, typically is employed to minimize the risk of conflict escalation. 9,[60][61][62] Anthropological field evidence demonstrates that reconciliation has a role in human populations as well. Among the Tsiman e people of lowland Bolivia, individuals prefer to solve their conflicts through reconciliation rather than revenge. 63 A reconciliation body, such as a council of elders, which presents another alternative, is used by the Pathan Hill tribes in Afghanistan. 64 Analogously, the most respected figures and leaders among the Ju/'hoansi Bushmen often refrain from taking part in the most severe punitive conflicts so that they can mediate reconciliation between the adversaries and prevent an escalation of aggression. 53 It seems clear that reconciliation is rather difficult without some sort of communication. One could argue that this behavior is greatly facilitated by the presence of language among humans. Nevertheless, since verbal or, generally speaking, personal interaction is typically unavailable in experiments investigating both costly and antisocial punishment, reconciliation is inaccessible. Consequently, punitive conflicts might needlessly escalate. Taking this into account, conducting a study in which both the opportunity of antisocial punishment and communication are enabled could simulate real-world conditions more accurately.

Punishment Outside of the Lab: The Many Faces of Costly Penalties
During the last fifteen years, intensive investigations of punishment, largely based on economic dilemma games, have clearly demonstrated that humans have a strong, universal motivation to punish cheaters in order to push them to cooperate. Nonetheless, as noted, the central issue with experiments investigating the relationship between punishment and cooperation in humans is their rigid nature. Punishment is often unrealistically considered as unconditional, uncoordinated, and directly provoked by defection. Since factors as characteristic of human societies as reputation and communication are commonly eliminated, the gravity of the problems typically associated with the costs of punishment is, perhaps, exaggerated.
In spite of all that, there remain some limitations on the power of punishment in the maintenance of cooperation. Clearly, punishment is a potentially dangerous tool, breeding aggression and cycles of retaliation that could lead to disarray, violence, or social breakdown. However, drawing on indirect reciprocity, coordination, and verbal reproach, humans have devised mechanisms to tame and control the hazardous nature of punishment and forged it into less destructive forms. As a result, punishment in real-world conditions is typically coordinated, associated with reputation, mediated by language and, briefly, less detrimental.

Indirect Reciprocity: When Punishment and Reputation Pull Together
"A good name is to be chosen rather than great riches, and favour is better than silver or gold." Proverbs 22:1 The main currency that indirect reciprocity brings into play is represented by reputation. One does not need to consult academic literature to figure out that reputation greatly matters in our social interactions and is relevant to punishment and cooperation: We despise cheaters and are more likely to punish in the presence of larger audiences. 18,65 Similarly, if reputation is allowed to form during experiments, general group-beneficial behavior, such as cooperation, is positively affected. 66,67 It appears also that the neurocognitive system involved in processing cooperation and punishment overlaps with the one assessing reputation. Merely observing the face of a partner with the reputation of a cooperator activates the same reward-related areas in the brain as punishment and cooperation do. 68,69 Looking at the neurocognitive basis of evaluating the faces of cheaters could be also insightful.
There seem to be at least two distinct ways that punishment and reputation can interact to promote cooperation. First, punishment can directly increase the reputation of punishers, making them more likely to receive cooperation in subsequent interactions as well as turning them into more desirable mates or allies. 70 Second, indirect reciprocity allows an individual to select cooperative partners and shun uncooperative ones based on their reputations; it thus provides an additional incentive to cooperate and reduces the need to use costly penalties. 51 Since cheaters suffer from low regard, cooperators might selectively withhold their cooperation from them and direct it toward individuals who are reputable cooperators instead. In consequence, as harmful interactions with free-riders are avoided, punishment need not be performed that frequently, which minimizes its overall costs and its negative effect on the efficiency of cooperation. Whereas the first option could be termed "building the reputation of a punisher," the second one might be called "building the reputation of a cooperator." Do we, therefore, build the former, the latter, or both?

Building the Reputation of a Punisher
It has been suggested that punishers gain an indirect benefit in terms of reputation, which directly compensates for the costs of punishment and thus could equally alleviate the second-order free-riding problem. 42,71 In this respect, reputable punishers should be less likely to be cheated or denied cooperation in future interactions and, at the same time, more likely to be regarded as desirable social and mating partners. But do punishers actually accrue benefit from their higher reputation?
The answer to this question is affirmative. In a computer public goods game, where the roles of punisher, free-rider, and nonpunisher were predetermined, players rated the former as more trustworthy, group-focused, and worthy of respect than the latter two. 72 Moreover, in subsequent iterated trust games, where participants were paired with one of the preprogrammed players, punishers enjoyed the highest trust and amassed highest earnings. Nevertheless, participants were sensitive to the fairness of punishment, as only justifiable cases of imposing penalties on free-riders were rewarded with increased reputation. There is evidence of a similar process in a real-world population. Among the Bergdama, a seminomadic people of Namibia, especially costly cases of punishment are usually carried out by younger men who, if successful, benefit from increased prestige. 73 Nonetheless, punishment can sometimes become too much of a good thing and prove detrimental to reputation. Among the Ju/'hoansi Bushmen, individuals who engage in overly frequent or harsh punishment are regarded as inconsiderate and do not benefit from high reputation. 53 Most esteemed figures or leaders, accordingly, refrain from taking part in severe punitive conflicts and draw their reputation from subsequent mediation rather than punishment. By the same token, when managers in companies are observed punishing their subordinates, their employees sometimes report that they are less trustworthy and respectful. 74 The relationship between punishment and reputation is therefore far from straightforward. It is clear that to a certain extent it pays off to be a punisher, in terms of both reputation and future interactions. Unjustified, overly harsh, or antisocial punishment, on the other hand, appears to damage a punisher's reputation. As stated, punishment need not always be the most preferable strategy in terms of reputation. Its appropriateness seems to depend on the context of a situation. The specifics of the link between reputation and punishment should become a priority in future research.

Building the Reputation of a Cooperator
Indirect reciprocity equally allows one to build the reputation of a cooperator and interact with other reputable cooperators while shunning cheaters. 24,49,75,76 By withdrawing cooperation, one can easily terminate or avoid harmful interactions at little cost. Defectors, on the other hand, are "punished" through social rejection or loss of help and mating opportunities. As cheaters are cut off from the cooperative circles, the detrimental effect of freeriding on these circles is mitigated and cooperation can thrive. In addition, as encounters with free-riders are intentionally avoided, punishers need not often inflict costly penalties, which mitigates the negative effect of punishment on the efficiency of cooperation.
An experimental study by Rockenbach and Milinski 59 allowed an interaction between withdrawal of cooperation and costly punishment. As expected, only the combination of both strategies yielded the most effective and efficient cooperation. Since participants were naturally concerned with building the reputation of a cooperator, indirect reciprocity provided an alternative incentive to cooperate, which in turn reduced the need for extensive use of costly penalties in order to enforce cooperation. Availability of the option to punish and discipline occasional defectors, however, was required if this effective and efficient cooperation was to be sustained. In the absence of costly penalties, when cooperation relied solely on reputation, its efficiency and effectiveness declined.
The strategy of withdrawing cooperation can draw on rich anthropological evidence. In the Pathan Hill tribes of Afghanistan, recurrent cheaters suffer from covert social shunning: for example, they are avoided in forming alliances and never invited to participate in the Council of Elders. 64 Similarly, among the Ju/'hoansi Bushmen, stingy or uncooperative individuals are held in low social regard, and thus have fewer marriage or social exchange opportunities. 53 The importance of reputation for finding a mate is demonstrated by a study in the Turkana, a nomadic society of Kenya, where, when asked about their feelings toward a fictional cowardly warrior, unmarried women tended to be more critical than any other respondents. 77 It seems clear that indirect reciprocity and costly punishment mutually reinforce cooperation. Although punishment can become a secondary measure under the rule of indirect reciprocity, its presence and occasional infliction are required if cooperation is to be sustained at a high level. To conclude and return to the initial question, humans conceivably build both the reputation of a cooperator and a punisher, since both bring benefits. The relative importance of the two should be investigated by future research.

Reputation Is Not Everything
Indirect reciprocity appears to be a very important mechanism in the maintenance of human cooperation. But what if information about repu-tation is unavailable or unreliable? Presumably, this is not an uncommon scenario in the real world. In this respect, as population size increases, information quality of systems based on reputation rapidly declines, thus making indirect reciprocity a doubtful candidate for sustaining cooperation at large scales. 51,77,78 In consequence, indirect reciprocity appears to bear more relevance for cooperation in the circle of close acquaintances for whom reputation is either available or can be easily acquired. To sustain cooperation beyond this circle, however, additional mechanisms, such as coordinated punishment and institutions, might be required.

Coordination: United We Stand, Divided We Fall
As the evidence from studies allowing endogenous decisionmaking demonstrates, if players are permitted to communicate and determine their own rules, cooperation thrives. 33,44,46,47 In this respect, group control and coordination of penalties bring numerous benefits. In the first place, coordination helps decrease the costs of punishment, which are either distributed among the whole community or managed in the cheapest way possible. 36,41,53 Moreover, if a rule is unified and backed by the group's majority or dominant members, its legitimacy is established, which reduces the likelihood of antisocial punishment. 79 Collective punishment thus provides the necessary control for the potentially detrimental effects of costly penalties, minimizing both their costs and the occurrence of antisocial punishment. This could equally help alleviate the second-order freeriding problem, since everyone may be expected to participate and share the costs or, alternatively, punishers can be appointed by the community.
As ethnographical studies indicate, punishment is rarely unconditional and uncoordinated in real-world populations.
The Turkana, a nomadic people of Kenya, sustain large-scale cooperation in combat through coordinated costly punishment. 77,80 Acts of free-riding such as desertion and cowardice are first discussed by the whole community, whose opinion then determines whether corporal punishment is appropriate. If so, its execution is administered by members of the violator's age group, thus not necessarily by those harmed by the cheating behavior. Such third-party action bears costs in terms of time and energy, and may damage social relationships, yet the communal consent and coordination ensure legitimacy and efficacy.
Control imposed by the community seems to be essential for the correct function of costly punishment. In the South African tribes of Bergdama and !Kung, as well as among indigenous Australians, serious transgressions of rules are broadly debated before a final decision is reached and suitable punishers nominated. 73,81,82 To prevent vengeful punishment, individuals responsible for mediation may be appointed or, alternatively, the act of punishment may be delegated to the violator's kin. 82 Coordination seems especially relevant to the most costly forms of punishment. Among the Ju/'hoansi Bushmen, the vast majority of mild disapproval is administered by individuals, but sharp criticism is mostly delivered by alliances. 53 Similarly, safety in numbers is more important when punishing particularly corrupt behavior that immediately threatens community stability.
In brief, anthropological field evidence suggests that human cooperation at larger scales, where reputation gradually loses its power, is to a great extent enforced by thirdparty coordinated punishment, especially as regards highly costly penalties or particularly dangerous forms of cheating. Although coordinated punishment still bears its costs, these are greatly reduced by the power of community. In societies that operate on extremely large scales, coordination of penalties eventually gives rise to institutions that establish firm control over crime and punishment, legitimize rules and penalties in codes of laws, and employ professional policing forces, judges, mediators and others. 51 In other words, unrestricted costly penalties are explicitly prohibited and punishment is centralized in the hands of the state, with cheaters or antisocial punishers actively prosecuted for their offences. Therefore, formation of institutions promoting cooperation in everyday interactions might have been especially crucial for the development of large-scale cooperation in humans. This possibility is supported by the fact that both market integration and the presence of a world religion positively co-vary with fairness across different cultures. 78 The Irresistible Urge to Gossip As ethnographic evidence indicates, punishment in real-world populations typically starts as talk and gradually escalates to more severe forms: from joking, mocking, and mild complaints to sharp criticism and eventually violent acts. 41,53,82 Wiessner's 53 investigation of punishment among the Ju/'hoansi Bushmen fittingly demonstrates the importance of verbal reproach. Out of 171 cases of punishment recorded, 98% were verbal (22% mocking and joking, 41% mild complaints, 35% sharp criticism) and only 2% of punishments were represented by costly violent acts. 53 Although it is debatable to what extent Wiessner's observations can be considered as punishment, she reports that among the Ju/'hoansi, verbal reproach, which is typically triggered by disruptive behavior and idleness, can bear substantial costs for both the actor and the recipient and, in many cases, leads to a change in the recipient's behavior -which fits the definition of punishment.
One specific form of verbal punishment has received considerably more attention than any other: gossip. It is no secret that chitchat is central to our nature: it has been estimated that gossip could account for 65% of all human conversation. 83 Through sharing diverse information about an absent third party and combining reputation and coordination, gossip enables fast and cheap communication of news across the whole community. 84 This constant flow of reputational information allows close surveillance of the behavior of others. Because of its particularly strong effect on reputation, gossip can promote cooperation in a number of ways. [83][84][85] First, by spreading reputational information about cheaters who are then mocked or shunned, gossip mediates the process of building the reputation of a cooperator and enables one to selectively withhold cooperation from defectors. Second, by increasing reputational incentives, gossip provides extra motivation to build the reputation of being a cooperator and refrain from cheating. In this regard, it has been found that in the presence of a bystander who, as conveyed to participants, was likely to gossip about the behavior observed, cooperativeness in economic games greatly increased. [83][84][85] Third, by showing that one is willing to spread negative information if cheated, gossip not only deters others from defecting, but also contributes to formation of the reputation of a punisher. While gossip can be equally abused to spread false information, liars and overly negative or frequent gossipers are likely to be distrusted and become targets of gossip themselves. 86 Although verbal punishment is typically regarded as bearing little or no price and thus as not being a strategy associated with punishment, this perspective greatly neglects the impact of social and psychological costs. In this respect, the outcome of overheard gossip in terms of damaged social relationships can be quite severe. 86,87 Similarly, among the Ju/'hoansi Bushmen, sharp criticism typically involved emotional pain and even material consequences for both the punishing and the punished individual. Very frequently, an open tension between the two remained and they would not engage in any sort of social exchange until the conflict was solved. 53 Yet despite the well-known risks, the Ju/'hoansi are not afraid to verbally express their disagreement.
In addition to costs, one more clue relates gossip to costly punishment. The negative affective response that drives punishment seems to be equally involved in encouraging gossiping behavior, which suggests that gossip is underpinned by the same motivation that underlies costly punishment in economic games. 84 Therefore, in light of the evidence presented, it appears that verbal reproach has more in common with punishment than one might initially say. It is frequently triggered by inappropriate behavior, can be rather costly, can elicit a change in behavior, and is probably driven by identical motivations.

Language Is Behind Everything
As argued here, in real-world conditions the motivation to punish can be manifested in many different, often cheaper and safer forms, since costly penalties can wear the mask of reputation, coordination, or verbal reproach. It can be argued that language, in particular, underlies these three factors and might therefore have made a highly significant contribution to the diversification and versatility of punishment in humans. 4,88 In this regard, the ability to speak greatly facilitates coordination, as it allows groups to define specific rules and laws. Moreover, the ease and speed of humans' verbal communication permitted a more efficient exchange of social information, for example through gossip, and thus allowed reputation to play a considerably more important role in the life of community. Finally, the immediate cost of both detecting and punishing cheaters is considerably lower if verbal communication and reproach are available. In brief, because of its low price and ability to mobilize collective action, language might be one of the key reasons behind the high success of punitive strategies among humans. It may equally reinforce other elements that promote cooperation among humans, such reputational mechanisms or signaling. 88 Since human cooperation occurs in a wide array of ecologies and domains, slightly different measures of control in terms of punitive behaviors might have been favored by selection in different environments. 77,89 Consequently, in conditions where reputation, coordination, or language provide cheaper options of social control, other, more costly means of punishment may become limited and rarely used. 59,77 The extraordinary diversity of punitive measures observed in the field testifies to this notion. Not only have real-world human populations presumably reached punishment equilibrium a long time ago, but they also had millennia to devise strategies that reduce the costs of punishment and tame its risks. 53,77 Among the Ju/'hoansi Bushmen, for example, only 2% of the penalties observed were responses to extremely costly, violent behavior. 53 In view of that, it is crucial to bear in mind that punishment need not always be extremely costly to promote cooperation, but it is precisely this eventual option of harsh penalties that backs up the use of milder forms of punishment. 90 To illustrate this point, metaphorically speaking, one could imagine costly punishment as a policeman with a gun. Merely his presence is usually sufficient to keep the peace and order, and even when the law is broken, he typically does not need to reach immediately for his gun. Instead, he can start with verbal reproach or threaten to use his weapon before he resorts to highly costly forms of punishment. Even though the gun represents an extreme and a rarely used measure, it grants the policeman additional power and provides him with strong leverage in case it is needed. In its absence, the policeman would lose these advantages, making his job of enforcing rules and maintaining obedience to them considerably more difficult. Therefore, although milder penalties will be used more frequently than costly punishment, the presence of this severe threat is what ultimately sustains their efficiency. As Hobbes wrote in his Leviathan, ch. XVII "Covenants, without the Box 1. Glossary

Actorfocal individual who performs a given behavior
Free-ridersee cheater Altruisma behavior that is costly* to the actor and beneficial to the recipient Gossipa behavior whereby individuals share evaluative information about an absent third party Antisocial punishmentpunishment directed at cooperators or punishers of cheaters Cooperationa behavior that provides a benefit to its recipient, and is selected because of this benefit Cooperation effectivenessthe extent to which cooperation can be sustained (usually in terms of contributions to the public pool) Cooperation efficiencythe extent to which cooperation pays off (usually in terms of net benefits) Cheateran individual who does not cooperate or cooperates less than a fair share and thus gains a benefit at the expense of cooperators Equilibriumthe stage at which punishment is established as a shared expectation Punishmenta behavior that is costly to both the actor and the recipient, and promotes the recipient's cooperation in subsequent interactions Recipientan individual who is affected by the behavior of the focal individual Second-order free-ridersindividuals who cooperate in the main activity but fail to punish free-riders, thus achieving higher net benefits than do punishers Selfishnessa behavior that is beneficial to the actor and costly to the recipient Spitea behavior that is costly to both the actor and the recipient *Costs and benefits are defined in terms of fitness.
Sword, are but Words, and of no strength to secure man at all".

CONCLUSION
The last decade of intensive research, predominantly drawing on economic games, has produced strong evidence that people across all cultures readily punish free-riders in order to push them toward cooperation, even if this behavior bears substantial costs. Indeed, the universal presence of a strong motivation to punish cheaters can be regarded as the most important piece of evidence offered by previous empirical efforts. New insights, nevertheless, have gone hand in hand with challenges to the notion that punishment could play a role in the maintenance of human cooperation. It was regarded as too costly and destructive, and as suffering from the problems of second-order free-riding and antisocial punishment.
However, as argued in the first section of this paper, the gravity of the suggested limitations may have been overestimated in early studies as a result of the features of experimental design. Most importantly, the role of factors as characteristic of human societies as reputation and language has been neglected. In the absence of communication, common expectations have to be formed through trial and error, which might underrate the efficiency of punishment, while the problems of secondorder free-riding and antisocial punishment may have been exaggerated due to inappropriate payoff matrices and anonymity, which protects deviant individuals. In the second part of the article, we have demonstrated how the combination of language and indirect reciprocity allowed humans to shape costly punishment into numerous low-cost and less detrimental strategies.
Although the combination of communication and reputation might seem to be able to alleviate most problems associated with costly penalties, strong and precise experimental evidence for some of the claims introduced here is currently missing. Most notably, the specifics of the relationship between punishment and indirect reciprocity -for example, the extent to which punishers benefit from improved reputationhave not been satisfactorily addressed. Second-order free-riding also deserves further consideration: Has the problem been exaggerated by the use of inappropriate payoff matrices disregarding reputation, for example, or did humans evolve a different solution, perhaps in terms of coordination? Moreover, researchers have yet to shed light on the role of anonymity in antisocial punishment, the importance of reconciliation, the construction of common expectations, and the achievement of equilibrium.
In spite of having provided priceless insights, models and experimental games can only take us so far. Without a change in direction, we cannot expect new findings to emerge. Above all, future research should aim to expand on the previous restrictive experimental design and prioritize investigations of punishment in more natural settings, where the influence of and interaction with reputation, communication, and possibly other factors could be studied. Field studies especially have the potential to become a decisive source of data and new knowledge. Only this approach can lead to a more complete picture of costly punishment and its role in the maintenance of large-scale cooperation in humans.