Chapter 1, The Problem of Cooperation. Why do people (or other actors) cooperate? "The objective of this enterprise is to develop a theory of cooperation that can be used to discover what is necessary for cooperation to emerge." It uses the Prisoner's Dilemma as a framework for testing theories about balancing self-interest and competition.

"In the Prisoners' Dilemma, the strategy that works best depends directly on what strategy the other player is using and, in particular, on whether this strategy leaves room for the development of mutual cooperation."

Chapter 2, TIT FOR TAT. "The iterated Prisoners' Dilemma has become the E. Coli of social psychology," yet people have not paid much attention to how to play the game well. Axelrod organized a computer tournament to which people familiar with PD submitted programs encoding different strategies. The winner was one of the simplest, TIT FOR TAT.

Axelrod then constructed an environment in which different programs competed, and the losing programs were eliminated: this was an ecology that rewarded high scoring programs, and punished others. "This process simulates survival of the fittest. A rule that is successful on average with the current distribution of rules in the population will become an even larger proportion of the environment of the other rules in the next generation. At first, a rule that is successful with all sorts of rules will proliferate, but later as the unsuccessful rules disappear, success requires good performance with other successful rules." In other words, the competition gets tougher.

"The analysis of the tournament results indicate that there is a lot to be learned about coping in an environment of mutual power. Even expert strategists from political science, sociology, economics, psychology, and mathematics made the systematic errors of being too competitive for their own good, not being forgiving enough, and being too pessimistic about the responsiveness of the other side."

The tournaments reveal that "there is a single property which distinguishes the relatively high-scoring entries from the relatively low-scoring entries. This is the property of being nice, which is to say never being the first to defect."

TIT FOR TAT's rules for success:

• Be nice. Don't be the first to go on the attack. This demonstrates good will, and avoids provoking others.
• Retaliate. If others attack, retaliate. Not doing so encourages bad behavior and gives niceness a bad reputation.
• Be forgiving. If others defect but then go back to cooperating, accept the opportunity to move back to a cooperative mode.
• Be clear. Others can predict what you'll do, be certain that their moves will have definite outcomes. "There is an important contrast between a zero-sum game like chess and a non-zero-sum game like the iterated PD. In chess, it is useful to keep the other player guessing about your intentions. The more the other player is in doubt, the less efficient will be his or her strategy. But in a non-zero-sum setting it does not always pay to be so clever. In the iterate PD, you benefit from the other player's cooperation."

Chapter 4, Trench Warfare. During World War I, "live and let live" arrangements emerged spontaneously between opposing units on the Western Front. Cooperation could take hold because "the same small units faced each other in immobile sectors for extended periods of time." Consequently, they had a more sustained relationship than in mobile warfare, and could develop commonly-understood rules, reciprocity and restraint in attacks, displays of strength (e.g., snipers shooting at hard targets)as well as ethics (recognition that there was an arrangement and violating it was immoral) and rituals (e.g., regular artillery firing).

"Cooperation first emerged spontaneously in a variety of contexts, such as restraint in attacking the distribution of enemy rations, a pause during the first Christmas in the trenches, and a slow resumption of fighting after bad weather made sustained combat almost impossible. These restraints quickly evolved into clear patterns of mutually understood behavior, such as two-for-one or three-for-one retaliation for actions that were taken to be unacceptable."

Chapter 6, How to Choose Effectively. Four suggestions about how to do well in PD:

• Don't be envious. In a PD, "envy is self-destructive. Asking how well you are doing compared to how well the other player is doing is not a good standard unless your goal is to destroy the other player." However, in an iterated prisoner's dilemma, you can't do better than the other player, unless they're always suckers. "In a non-zero-sum world you do not have to do better than the other player to do well for yourself. The other's success is virtually a prerequisite of your doing well for yourself."
• Don't be the first to defect (be nice). "It pays to cooperate as long as the other player is cooperating." In a short game, defection can make sense; but in a relationship, taking advantage of the other person is self-defeating.
• Reciprocate both cooperation and defection. TIT FOR TAT "does not destroy the basis of its own success. On the contrary, it thrives on interactions with other successful rules." However, the right level of forgiveness depends on the context, and the other players' strategies.
• Don't be too clever. "In a zero-sum game, such as chess it pays for us to be as sophisticated and as complex in our analysis as we can. Non-zero-sum games are not like this. The other player can respond to your own choices. And unlike the chess opponent, the other player in a PD should not be regarded as someone who is out to defeat you." "There is an important contrast between a zero-sum game like chess and a non-zero-sum game like the iterated PD. In chess, it is useful to keep the other player guessing about your intentions. The more the other player is in doubt, the less efficient will be his or her strategy. But in a non-zero-sum setting it does not always pay to be so clever. In the iterate PD, you benefit from the other player's cooperation."

Chapter 7, How to Promote Cooperation. Promoting cooperation can be thought of as an exercise in tinkering with the variables in a PD. "As long as the interaction is not iterated, cooperation is very difficult. That is why an important way to promote cooperation is to arrange that the same two individuals will meet each other again, be able to recognize each other from the past, and to recall how the other has behaved until now."

• Enlarge the shadow of the future. For cooperation to emerge, players must be in a continuing relationship, with the expectation that it will continue in the future. "Mutual cooperation can be stable if the future is sufficiently important relative to the past." "There are two basic ways of doing this: by making the interactions more durable, and by making them more frequent. [P]rolonged interaction allows patterns of cooperation which are based on reciprocity to be worth trying and allows them to become established," Making interactions more frequent makes "the next interaction occur sooner, and hence the next move looms larger than it otherwise would." You might do this by enforcing isolation, or constructing hierarchies or organizations, which are "especially effective at concentrating the interactions between specific individuals."
• Change the payoffs. Make defection less attractive, by enforcing laws, or growing the value of long-term incentives.
• Teach people to care about each other.
• Teach reciprocity. Reciprocity "actually helps not only oneself, but others as well. It helps others by making it hard for exploitative strategies to survive."
• Improve recognition abilities. "The ability to recognize the other player from past interactions, and to remember the relevant features of those interactions, is necessary to sustain cooperation. Without these abilities, a player could not use any form of reciprocity and hence could not encourage the other to cooperate."

Chapter 8, The Social Structure of Cooperation.
The social structure of cooperation involves labels, reputation, regulation, and territoriality.

• Labels are fixed characteristics of an agent that are observable by other agents. Labels affect reciprocity and retaliation via assumptions of group similarity and stereotypes.
• Reputation is others' belief about the strategies an agent will employ. Reputation may be based on past behavior or on rumours, i.e. reputation can be accurate or merely believed. Reputation affects whether or not other agents will cooperate or defect with you.
• Regulation involves setting the stringency of a standard of behavior "high enough to get most of the social benefits of regulation, and not so high as to prevent the evolution of a stable pattern of voluntary compliance from almost all of the companies" (or regulated agents).
• Territoriality refers to both physical and conceptual spaces that can be 'invaded' by agents of differing strategies. Territoriality establishes boundaries within which behaviors will be reinforced or retaliated against depending on prevailing norms. Also, the boundary provides an 'inside' for agents that comply with the norms, and an 'outside' to which they can be expelled if they do not comply.

Chapter 9, The Robustness of Reciprocity.

• Cooperation can get started by even a small cluster of individuals who are willing to reciprocate cooperation, even in a world where no one else will cooperate.
• Once cooperation is establish, it protects itself from invasion by non-cooperative strategies.
• The foundation of cooperation is the durability of the relationship, which allows agents to learn about each other in order to cooperate.

“The study of social dilemmas is the study of the tension between individual and collective rationality. In a social dilemma, individually reasonable behavior leads to a situation in which everyone is worse off. The first part of this review is a discussion of categories of social dilemmas and how they are modeled.” The Prisoner’s Dilemma, the problem of providing public goods, and Hardin’s Tragedy of the Commons are three powerful metaphors that facilitated and structured research but also served as blinders since their limitations are often not recognized.

Models:

Kollock’s analysis divides dilemmas into two-person and N-person dilemmas. The key two-person dilemmas are the Prisoner’s Dilemma, the Assurance Game, and the Chicken Game. Each of these models is defined by the ordering of four possible outcomes: mutual cooperation, mutual defection, and either first or second person’s unilateral defection. Each of these outcomes generates an individual benefit for each person and is ordered by the benefit for the first person.

The Prisoner’s Dilemma models unsecured transactions, e.g. buying and selling over the Internet. The best outcome of a Prisoner’s Dilemma is unilateral defection of the first person, followed by mutual cooperation, mutual defection, and the worst outcome is the first person’s unilateral cooperation. Since defection has the highest potential benefit and cooperation the highest potential risk, the equilibrium of the Prisoner’s Dilemma is mutual defection. This equilibrium is deficient because the best outcome for both players is mutual cooperation.

The Assurance Game is similar to the Prisoner’s Dilemma except it models situations where mutual cooperation is more benefical for each player than unilateral defection, e.g. a project that requires collaboration. This extra motivation to mutually cooperate creates two equilibria, one optimal, which is mutual cooperation, and one deficient, which is mutual defection. The optimal equilibrium requires trust between the two persons sufficient to assure each other that the other will cooperate. Insufficient trust leads to the deficient equilibrium.

The Chicken Game is again similar to the Prisoner's Dilemma except mutual defection is the worst outcome, worse than unilateral cooperation. This replaces the Prisoner’s Dilemma’s mutual defection equilibrium by two equilibria, unilateral defection and unilateral cooperation because of the strong motivation to not mutually defect. The Chicken Game is a model for situations that require volunteer effort to avoid the worst outcome but where duplicate effort is less desirable.

Kollock divides N-person dilemmas into two types based on cost and benefit for each individual. The first type is known as the social fence,s where an individual is presented with an immediate cost that generates a benefit shared by all. The individual wants to avoid the cost but if all do, everyone is worse off. A common metaphor of the social fence is the provisioning of public goods, which are (to a varying degree) non-excludable and nonrival. The key characteristic of a public good dilemma is the production function which defines the relationship between the level of resources contributed and the level of public good provided. Production functions are classified into decelarating, linear, accelerating, and step functions. Various production functions can produce N-person versions of any of the 2-person dilemmas.

The second type is know as social trap where the “individual is tempted by an immediate benefit that produces a cost to all. If all succumb to the temptation, the outcome is a collective disaster.” The usual metaphor of the social trap is the tragedy of the commons. A key feature of commons dilemmas is that the benefits are non-excludable (or difficult to make excludable) and subtractable. The key characteristic of commons dilemmas is the carrying capacity of the commons which depends on the replenishment rate of the subtractable joint resource.

Important (but not inevitable) features that affect N-person dilemma dynamics and contrast them to two-person dilemmas are anonymity, diffusion of defection cost, and little or no direct control on others. Some of these features are also found in two-person dilemmas, e.g. blaming defection on out-of-control circumstances is a form of anonymity in two-person games.

Solutions:

“The second part of [Kollock’s paper] is an extended treatment of possible solutions for social dilemmas. These solutions are organized into three broad categories based on whether the solutions assume egoistic actors and whether the structure of the situation can be changed: Motivational solutions assume actors are not completely egoistic and so give some weight to the outcomes of their partners. Strategic solutions assume egoistic actors, and neither of these categories of solutions involve changing the fundamental structure of the situation. Solutions that do involve changing the rules of the game are [called] structural solutions.”

The motivation of not completely egoistic actors to cooperate is influenced by social value orientation, communication, and group identity. The social value orientation of a person seems to be acquired from the person’s social environment and is some linear combination of a cooperator who tries to maximize joint outcome, a competitor who tries to maximize own outcome relative to partner, and an individualist who tries to maximize own outcome. Kollock does not find any conclusive results in how to influence social value orientation but does find evidence that it varies between different countries.

The presence of communication positively affects cooperation rates. Communication enables a person to find out about others’ choices, to make explicit commitments, to appeal to what is the moral thing to do, and most importantly, to create or reinforce a sense of group identity. The effect of group identity is in fact so strong that it can affect cooperation rates even in the absence of communication. In-group behavior of individuals frequently includes personal restraint and treating Prisoner’s Dilemma situations as Assurance Games. However, in-group behavior implies out-group behavior with the potential to cause severe social costs due to intergroup conflicts.

“[Strategic solutions] rely on the ability of [egoistic] actors to shape to shape the outcomes and hence behavior of other actors. For this reason, many of these strategic solutions are limited to repeated two-person dilemmas.” Axelrod (see The Evolution of Cooperation) identifies three requirements for strategic solutions: ongoing relationships between actors (i.e. all expect shared dilemmas in their future), ability to identify each other, and ability to keep track of the other’s past behavior.

The most successful strategy in iterative Prisoner’s Dilemma tournaments (everyone against everyone) that meet these requirements is Tit-for-Tat which starts out with cooperation and then matches the partner’s previous behavior. This strategy transforms a repeated Prisoner’s Dilemma into a repeated Assurance Game since the only long-term outcome of this strategy is either mutual cooperation or mutual defection (the two equilibria of the Assurance Game). Key aspects of successful strategies in repeated Prisoner’s Dilemma tournaments are (1) to realize that it is not a zero-sum game hence does not benefit from a competitive social orientation (“don’t be envious”), (2) to not defect first, (3) to reciprocate both cooperation and defection, and (4) to be predictable so that the partner clearly understands one's strategy. One important caveat is that repeated Prisoner’s Dilemma tournaments assume perfect communication. In real life where communication is often imperfect more generous or forgiving strategies can avoid accidental cycles of recrimination.

Recent evidence suggests that the strategy of choosing partners is more important than the strategy used within a dilemma. In a modified version of iterative Prisoner’s Dilemma tournament actors can exit current relationships and choose alternative partners. A very successful strategy in this environment is Out-for-Tat which exits a relationship as soon as the partner defects. A more forgiving version that gives a defecting partner a second chance is even more successful.

Strategies for N-person dilemmas involve grim triggers, social learning, and group reciprocity. In a “grim trigger” strategy an individual only cooperates if all other group members cooperate and defects as soon as one other group member defects. Social learning is the basis of a cognitively less taxing class of strategies that involves imitating other group members and look for thresholds in public good provisioning instead of calculating marginal rates of return or figuring out dominating strategies. Group identity increases cooperation rates because group members follow strategies that assume that all members share a strong expectation of group reciprocity (reciprocity within the group).

Structural solutions change the rules of the dilemma thereby changing or eliminating it. One approach is to reinforce prerequisites for strategic solutions by introducing long-term accountability (shadow of the future) that influences individual reputations. However, accountability and reputation are not sufficient to escape the Prisoner’s Dilemma’s equilibrium of mutual defection (in two- or N-person version) if the means to encourage cooperation are too weak (e.g. production function for public good too flat or too much effort required to reach provisioning point).

Many people seem to positively weigh others’ outcomes since cooperation increases significantly as the benefits to others from one’s cooperation increase. Cooperation levels are also higher if group members are asked to contribute to a non-divisible public good that only benefits the whole group, probably due to an increased sense of group identity (see group reciprocity).

Cooperation in N-person dilemmas increases if individual contributions have (or are perceived to have) a discernable effect, i.e. make an efficacious contribution. For public goods with step-level production function one can create a minimal subgroup that requires every member to contribute in order to reach the provisioning point or let two groups compete for contributions, turning an N-person Prisoner’s Dilemma into an N-person Chicken Game. Another example are "matching grants" or "adopting" an individual from a large group of benefactors.

Increasing group size makes defection more anonymous and increases the cost of organizing. However, research results on cooperation depending on group size alone are inconclusive. In the case of highly non-rival goods with a threshold production function a larger group is more likely to contain a "critical mass" of cooperating individuals. Diversity of group members' interests and resources encourages formation of critical mass.

A common structural strategy for N-person dilemmas is the creation of boundaries in an attempt to make public goods or commons more excludable. There are three main approaches: The first one is to institute an external authority or trusted leader to govern access to commons. This approach appears to be less preferable if other structural changes are possible. Establishing an external authority can raise severe problems of justice, enforcement, corruption, and scalability. The second approach is to break up commons into private parcels assuming that individuals will take better care of own property than common property. However, privatization does not work for non-divisible goods, raises the social question of who gets to own commons, does not prevent owners to routinely destroy their own property (“tragedy of enclosure”), and requires institutional support to enforce private property rights. A third approach is to locally regulate “access to and use of common property by those who actually use and have local knowledge of the resource.” One key characteristic of successful and long-lasting local regulations is clearly defined boundaries.

Sanctions are a structural method to encourage cooperation where the outcomes themselves of N-person dilemmas are too weak of a motivator. However, the implementation of sanctions can be very expensive. Local monitoring and sanctioning systems are more practical and less costly. Another way to reduce cost is to use a graduated system of sanctions with low-cost conflict resolution. A sanctioning system is itself a public good and therefore poses a second-order dilemma. Communities with a high level of trust readily cooperate in a first-order dilemma but cooperate less in a second-order dilemma hence are less willing to support a sanctioning system. The opposite is true for communities with a high level of distrust.

The self-regarding and outcome oriented picture of human behavior presented in traditional economics does not explain why humans care so much about each other and about how social interaction is carried out, not just the end goals. The Ultimatum Game, designed by Werner Guth, is just one illustration of how real people will not always follow the dictates of self-interested rationality. Two subjects are given a sum of money, one is given the power to divide the sum, and the other can either accept or reject (in which case neither get any money). Research from conducting hundreds of trials of the game with thousands of students in Europe, Japan and the USA has shown that the responders frequently reject low offers and proposers frequently propose near equal divisions, even though it is to their monetary disadvantage. While early experiments on undergraduates seemed to suggest that there was a universal sense of fairness, extended research in different cultures (hunter-gatherers, slash-and-burn agriculturists, nomadic pastoralists) has exposed much cultural variation in responses, indicating that local cultural conditions play an important role in how people approach cooperation.

While mean proposals for university students from all over the world was usually between 42 and 48 percent, mean proposals from this cross-cultural study varied from 25 to 57 percent. Rejection rates, the action of the responders, also varied considerable between groups. Individual-level economic and demographic variables did not explain behavior as well as group-level behavior, and game play often could be connected to the people's common patterns of interaction. For example, the Orma recognized that one of the experiment's games was similar to the harambee, a local institution of giving to public goods like roads and schools. They began calling it 'the harambee game' and displayed highly prosocial behavior. In other groups, like the Au and Gnau, frequent rejection of generous offers can be explained by a cultural association with gift-giving: accumulating gifts, even if unsolicited, can imply a lowered status and force the receiver into future obligations or political alliance. The cross-cultural study showed that, in the case of groups at the extremes of behavior, "contrasting behaviors seem to reflect their differing patterns of everyday life, not any underlying logic of hunter-gatherer life ways."

The effect of market integration on cooperation to obtain a monetary reward can be explained easily: individuals from market-oriented societies when put in the context of one of the games are able to seek analogues in their daily activities of using and trading money with strangers. "Those who do not customarily deal with strangers in mutually advantageous ways may be more likely to treat anonymous interactions as hostile or threatening, or as occasions for the opportunistic pursuit of self-interest."

The seven games explored here, ultimatum, public goods, dictator, prisoner's dilemma, trust, gift-exchange, and third party punishment, can be used both as metaphor to describe prototypical situations in the social world and as a tool to predict the behavior of players in the context of other players' likely actions. Data on the responses of real players can help guide the formation of successful and sustainable institutions for collective action. In a public goods game, for instance, contributions to the public good declined over repeated periods as cooperative players eventually became frustrated with an instigating group of free-riders. Once the structure of the game is altered to allow for punishment of free riders, the average contribution rises steeply to over 95 percent of the endowment. The actual rate of punishment does not have to be that high to generate this increase either; "the mere threat of punishment, and the memory of its sting from past punishments, is enough to induce potential free riders to cooperate."

Another alteration that increases cooperation is permitting communication. "Communication allows the conditional cooperators to coordinate on the cooperative outcome and it may also create a sense of group identity." In the trust game, an investor gives an amount to a trustee, which is tripled and the trustee can give any amount from all to nothing back to the investor. Positive reciprocity, a sense of obligation to repay trusting investors that arises in the trust game, is an important key to harnessing cooperation. Implicit social contracts built on the basis of positive reciprocity are cheaper to implement and can be more successful than explicit contracts.

The environment of our evolutionary adaptation can theoretically explain the origin of these preferences in repeated game settings. Evolution equips people with the cognitive ability to learn social norms and resulting strategies rather than having them hard-wired into the brain. This accords with the game theory conclusion that the best strategy depends on the structure of social relations and potential for norms to take hold and be effective.