In contrast to other examples of ultrasociality19–22 (for example, clones, bee hives or termite colonies), human cooperation is due less to kin selection than to cultural forces rooted in pervasive moral systems. From hunter tribes and village communities to nation states and global enterprises, the economic effects of nepotism, although certainly present, are minor compared with those of reciprocity.

Imagine a population of individuals with the option to help one another or not. Random pairs of players are chosen, of which one is the potential donor of some altruistic act and the other is the recipient. The donor can cooperate and help the recipient at a cost c to himself, in which case the recipient receives a benefit of value b (with b . c). If the donor decides not to help, both individuals receive zero pay-off. Each player has an image score, s, which is known to every other player. If a player is chosen as a donor and decides to cooperate then his (or her) image score increases by one unit; if the donor does not cooperate then it decreases by one unit. The image score of a recipient does not change.

Figure 1 shows computer simulations of a population consisting of n players. The strategies are given by ki and the image levels by si. At the beginning of each generation, the image levels of all players are zero (assuming that children do not inherit the image of their parents). In succession, m donor–recipient pairs are chosen. A donor, i, cooperates with a recipient, j, if ki # sj. The fitness of a player is given by the total number of points received during the m interactions. Some players may never be chosen, in which case their payoff from the game will be zero. On average, a player will be chosen 2m/n times, either as donor or as recipient. At the end of each generation, players leave offspring in proportion to the their fitness. We find that if the game is played for many generations, then eventually all players will adopt the same strategy. If the k value of this strategy is 0 or less then cooperation is established; if the value is 1 or more then defection has won. Cooperation is more likely to win if the number of interactions, m, per generation is large.

Builds on: "The image score of a recipient does not change."

A minimum number of rounds per generation is needed for cooperation to prevail. This number can be very small: each player needs to be chosen for only about two interactions per lifetime. (In this case, there is a probability of only 1/4 that a defector can be punished, if he is chosen first as a donor and then as a recipient.)

Builds on: "Frequent interactions per generation favor the evolution of cooperation."

There is also an effect of group size. For larger groups, it is more difficult to establish cooperation, because the fraction of individuals that obtain information about any particular interaction will be smaller. Therefore, more interactions are required (relative to group size) in order to discriminate against defectors.

Another interesting expansion of the basic model is to include strategies that consider both the recipient’s and the donor’s image score. We explored two types of strategies. ‘And’ strategies involve cooperation if the image score of the recipient is larger than a certain value and the image score of the donor is less than a certain value. The idea is that if an individual has already a high image score, it is not necessary to aim for a still higher image score (by helping others). On the other hand, ‘or’strategies result in cooperation if the image score of the recipient is larger than a certain value or the image score of the donor is less than a certain value. Here the idea is that if an individual has a low image score it may be advantageous to increase the score by helping others regardless of how low their image score is. In both cases, highly cooperative societies form (Fig 4). If, in contrast, we simulate strategies that only consider their own image and do not take into account the image of the recipient, cooperation does not emerge.

XREF: Connects to indirect reciprocity and reputation-based cooperation work in evolutionary game theory. Builds on: "A minimum number of rounds per generation is needed for cooperation to prevail."

The models above are based on computer simulations, but we can derive analytical insights from a simplified model. Suppose that there are only two image levels, 0 (for bad) and 1 (for good). The image of a player depends on his or her last action as a donor: players who defected have score 0, and players who cooperated have score 1. Let us only consider two types of player: first, defectors, who never provide assistance; and second, discriminators who help players having image 1, but not players having image 0. A given player knows the score of only a fraction, q, of the population.

DEFINE: Introduces the two-level (0/1) image score model and its player types: defectors and discriminators.

Consider two image scores, 0 for someone who defected last round and 1 for someone who cooperated last round. Thus the image score depends only on the last move of a player as a donor. Consider two types of players: discriminators, who help only players with image score 1, and defectors, who never help.

DEFINE: Introduces the image score model, a simple binary reputation system based only on a donor's most recent action, contrasted with reciprocal strategies and discriminators who help only the reputably cooperative.

Indirect reciprocity only works when donors discriminate between individuals that have or have not helped others in the past. To understand the role of indiscriminate altruists, we add to the population of defectors and discriminators a fraction z of cooperators, who always give help irrespective of the their co-player's score.

Selective representation of relevant information by neurons in the primate prefrontal cortex … Gregor Rainer, Wael F. Asaad & Earl K. Miller Department of Brain and Cognitive Sciences and The Center for Learning and Memory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA . . . The severe limitation of the capacity of working memory, the ability to store temporarily and manipulate information1, necessitates mechanisms that restrict access to it. Here we report tests to discover whether the activity of neurons in the prefrontal (PF) cortex, the putative neural correlate of working memory2–8, might reflect these mechanisms and preferentially represent behaviourally relevant information. Monkeys performed a ‘delayed-matching-to-sample’ task with an array of three objects. Only one of the objects in the array was relevant for task performance and the monkeys needed to find that object (the target) and remember its location. For many PF neurons, activity to physically identical arrays varied with the target location; the location of the non-target objects had little or no influence on activity. Information about the target location was present in activity as early as 140 ms after array onset. Also, information about which object was the target was reflected in the sustained activity of many PF neurons. These results suggest that the prefrontal cortex is involved in selecting and maintaining behaviourally relevant information.