Prisoner’s Dilemma teaches us about prediction

The Prisoner’s Dilemma game tests not only the rationality of the players but also their social decision-making skills. These skills are based on the player’s ability to predict what the other person will do. In the game two criminals, both aware of the greater award if they both cooperate, are most likely to betray each other because of each one’s prediction that the other will behave in their own self-interest.

Massachusetts General Hospital researchers led by Keren Haroush and Ziv Williams experimented with the Prisoner’s Dilemma game by applying it to Rhesus monkeys. In these experiments the researchers sought to identify neurons specifically involved in predicting another’s intended actions. Appearing in the March 12 issue of the journal Cell, this study delineates how two groups of neurons in the dorsal region of the anterior cingulate cortex (dACC) play a critical role in social decision-making.

During their investigation the researchers focused on the neural activity within the dACC, a region known to be involved in encoding social interest. To pinpoint the neurons concerned with predicting another’s intentions, the researchers tested pairs of Rhesus monkeys in a Prisoner’s Dilemma scenario. The animals were given varying amounts of juice as reward after either deciding to cooperate by picking an orange hexagon or deciding to defect by picking a blue triangle. The experiment was designed so that neither monkey could predict the other’s actions from facial expression cues or eye contact. Facing the screen the monkeys, though sitting side by side, could not observe what the other had chosen until after a blank screen delay.

If both chose to cooperate, they received equally large amounts of juice. If one of them had chosen to defect, the defector received a larger drink than the monkey who had decided to cooperate. If neither of them chose to cooperate, both received equally small portions.

As the monkeys were tested in multiple successive trials, their decisions to cooperate or defect were influenced by prior trials and observations concerning the other monkey’s choices. Generally the monkeys chose to defect (65.3 percent of the trials) rather than cooperate (34.7 percent). Moreover, if the monkeys had both cooperated on a previous trial, they were more likely to cooperate again, even though the defector would have received the highest individual reward, thus indicating their willingness to reciprocate a mutual benefit.

While the monkeys played the game in order to get the greatest amount of individual reward, the researchers noted the significant firing activity exhibited by 185 of the 363 dACC neurons that had been recorded. Many of these dACC neurons responded to the primate’s own decision, while a distinct neural population in the dACC seemed to anticipate the other monkey’s choice. Only 4.3 percent of the dACC neurons responded to both the monkey’s own decision and the opposing monkey’s anticipated selection.

The disparate dACC neurons that encoded the prediction of the other monkey’s selection were also influenced by the social context in which the primates played the game. To test whether the monkeys’ choices were influenced by the opposing player’s presence in the room, researchers set up the same Prisoner’s Dilemma trial, though this time assessing the monkey’s choices as it played against a computer. In another social control experiment, the two monkeys were placed in separate rooms.

In both of these controls, the monkeys were much less likely to cooperate on the whole, choosing the orange hexagon on 19.1 percent of the trials with the computer and on 14.2 percent of the trials when the monkeys were separated. Furthermore, the monkeys were less likely to reciprocate after having both cooperated on a previous trial.

In the control experiment involving the separate rooms, the investigators noted that a smaller fraction of dACC neurons exhibited significant activity in anticipation of the other monkey’s decision whereas a larger fraction of dACC neurons responded to the primate’s own selection.

The effect of the social environment on the primates’ performance in the Prisoners’ Dilemma game reveals the dynamic nature of the dACC neurons. While previous studies have assessed neurons that encode another’s observable actions, as in the case with canonical mirror neurons, this investigation has revealed a disparate neuronal population which anticipates the other individual’s actions.

The further exploration of the mechanism of how social anticipation is encoded within the brain will hopefully yield information that can be used to develop more efficient treatment for disorders, such as antisocial personality disorder and autism spectrum disorders, both of which are characterized by difficulty with social interaction and abnormalities in the dACC.