Pols 426 Lecture 9 R. Keohane and After Hegemony and The Simulation SimSociety

Several general patterns emerge from the simulation runs. First, as presented in Table 1, most simulation runs (83%) achieve a high level of cooperation (the average cooperation rate among agents is over 95%) at some point during the course of the simulation run. Fifty-nine (65%) simulation runs eventually end up in what appears to be a stable cooperative equilibrium (a high level of cooperation is achieved at some point in the simulation and is maintained until the end of the simulation) and sixteen simulation runs (18%) in a punctuated equilibrium (long periods of stable high levels of cooperation punctuated by periodic massive dips to near universal defection). Only 17% (15 runs) end up in a stable non-cooperative equilibrium (the average cooperation rate quickly declines and stays at less than 5% for the entire simulation run). ** Cooperation in an anarchic, self-help world where there is both opportunities for conflict and cooperation and where there is long terms interaction (shadow of the future) is quite likely to occur. But it is not guaranteed to either happen at all and if achieved to last; it can collapse -- So in Keohane’s terms cooperation without a hegemon is certainly possible

Second, the transition from near universal defection to near universal cooperation is always characterized by two features; networks of cooperative agents form and these networks are comprised of agents employing versions of the Grim strategy. *** Cooperation requires a primitive form of organization and community and a very tough stance toward non-cooperative behavior – Retaliation and non-forgiveness are important features - International regimes, in order to survive need some "mass" and organization and they must punish free riders and potential exploiters – All must punish exploiters in place of the hegemon and forgiveness at least initially cannot be tolerated without a hegemon

Third, high levels of cooperation characterized by large clusters of cooperative agents can collapse. When cooperation collapses, it is usually because the networks of Grim-like agents evolve into networks made up of agents with strategies that are "too nice." The typical pattern that leads to collapse is the upward drift in the probability of cooperating following a c,d outcome. In these cases, average c,d probabilities drift up to over .8. These agents, having a strategy mix such as [.99, .85, .10, .07], can easily be taken advantage of allowing nasty strategies to emerge and succeed. This leads to a typically long and slow (up to 100,000 iterations) decline to near total defection. Cooperation does not end quickly. It changes or decays very slowly – almost invisibly in short periods of time. The generation of too many nice strategies makes successful exploitation too easy leading to the collapse of cooperation. We can liken this situation to the "classic predicament of societies going ‘soft’ --- Nice agents like and succeed in a very cooperative world where almost all interaction is cooperative and they "forget" over time that they must punish defection --- particularly when they cooperated (c,d) but also after mutual defection (d,d) -- Pavlov does well in cooperative environments – indeed by being exploitative (defect after d,c), but because it does not punish mutual defection, it is itself subject to exploitation --- Regimes that have a norm of reciprocity and also one where defections are consistently sanctioned even when it is costly to the particular agent that must sanction are much more likely to last – maintain cooperation

We can assess whether groups of agents with various types of cooperative strategies can "invade" a set of All-D [0,0,0,0] agents. We start with 50 All-D agents randomly distributed on the grid, and 10 agents also randomly distributed on the grid for each of the following four types of cooperative strategies; initially "pure" TFT [1, 0, 1, 0], Pavlov [1, 0, 0, 1], Grim strategies [1, 0, 0, 0], and All-C [1,1,1,1]. Five simulations of each strategy mix (i.e., 50 All-D and 10 TFT) were run for both local and global movement where each simulation was run for 200,000 iterations

Not surprisingly, "invasion" is more successful with a set of Grim and TFT agents than with a set of Pavlov or All-C agents. The relatively very short time it takes a small number of initially pure TFT, or Grim strategies to dominate and create a cooperative world (an average of approximately 4,800 iterations) reinforces earlier observations. Non-cooperative worlds can be invaded by nice agents but they must be willing to retaliate. Forgiveness is ok (TFT) but greed coupled with openness to being exploited (Pavlov) is dangerous because it can prevent successful invasion and/or allow nasty agents to collapse an attained cooperative world.

More generally, Keohane argues that cooperation can be achieved and maintained by the presence of international regimes; principles, norms and rules guiding nation-state (agent) behavior. The various strategies of the iterated prisoners dilemma can be viewed as rule-based behavior and that norms or institutions of bilateral interaction may be represented as homogeneous systems of this rule based behavior. For an institution to exist in a population, it is not that all actor strategies are identical but rather that some behavioral characteristics are common to all members. Under these circumstances, these behaviors become norms and are institutionalized.

The results indicated that norms of niceness, and retaliation must be institutionalized if cooperation is to emerge and be maintained. Norms of nastiness do not generate cooperation – agents (nation-states) must be willing to initiate cooperation. Cooperation cannot be generated or maintained by institutions characterized by exploitation unless they also are nice. Cooperation cannot be generated or maintained by institutions characterized by niceness and forgiveness unless they are also coupled with retaliation.