Distinguishing the opponents in the prisoner dilemma in well-mixed populations

Here we study the effects of adopting different strategies against different opponent instead of adopting the same strategy against all of them in the prisoner dilemma structured in well-mixed populations. We consider an evolutionary process in which strategies that provide reproductive success are imitated and players replace one of their worst interactions by the new one. We set individuals in a well-mixed population so that network reciprocity effect is excluded and we analyze both synchronous and asynchronous updates. As a consequence of the replacement rule, we show that mutual cooperation is never destroyed and the initial fraction of mutual cooperation is a lower bound for the level of cooperation. We show by simulation and mean-field analysis that for synchronous update cooperation dominates while for asynchronous update only cooperations associated to the initial mutual cooperations are maintained. As a side effect of the replacement rule, an “implicit punishment” mechanism comes up in a way that exploitations are always neutralized providing evolutionary stability for cooperation.

💡 Research Summary

The paper investigates how allowing individuals to adopt different strategies against different opponents—rather than a single uniform strategy—affects the evolution of cooperation in the Prisoner’s Dilemma when the population is well‑mixed, i.e., without any network structure that could generate reciprocity effects. The authors introduce a novel “replacement rule”: when a player imitates a more successful strategy, the newly adopted strategy replaces the interaction (edge) that currently yields the lowest payoff for that player. Consequently, each pairwise interaction can maintain its own strategy, and any cooperation that ever emerges on a link can never be downgraded, making the initial fraction of mutual cooperation a strict lower bound for the eventual cooperation level.

Two updating schemes are compared. In synchronous updating, all agents evaluate and possibly replace their worst interaction simultaneously; in asynchronous updating, a single randomly chosen agent updates at each time step. The population is fully connected, ensuring that any observed effects stem from the replacement rule rather than spatial or network reciprocity.

Simulation results reveal a stark contrast between the two schemes. Under synchronous updating, cooperative strategies spread rapidly, eventually dominating the entire population. The simultaneous replacement of the worst links prevents exploitative (defector‑defector) ties from persisting, and any newly introduced cooperative strategy quickly overwrites them. Under asynchronous updating, however, the spread of new cooperative links is severely limited. Only the cooperative links that existed at the start survive; new cooperation rarely emerges because the stochastic, one‑by‑one replacement process does not provide enough opportunities for a cooperative strategy to replace a low‑payoff link before the system settles into a quasi‑steady state.

A mean‑field analytical framework corroborates these findings. The authors derive differential equations for the probabilities that a randomly chosen link is in a C–C, C–D, or D–D state. The replacement rule eliminates the C→D transition: once a link is C–C, it remains so because its payoff is never the lowest for either participant. Defector‑defector links are identified as the “worst” interactions and are preferentially targeted for replacement, which effectively implements an “implicit punishment” mechanism. Exploitative interactions are neutralized without any explicit cost or enforcement; defectors cannot reap long‑term benefits because any advantage they gain is quickly erased by the replacement of the exploited link.

The study highlights the evolutionary importance of strategy heterogeneity at the level of individual relationships. Traditional models assume a monomorphic strategy per individual, which often leads to the collapse of cooperation in well‑mixed settings. By allowing opponent‑specific strategies and a rule that protects any existing cooperation, the authors demonstrate that cooperation can be robust even in the absence of network structure.

Key conclusions are: (1) the initial proportion of mutual cooperation sets a lower bound for the final cooperation level; (2) synchronous updating drives the system toward full cooperation; (3) asynchronous updating preserves only the initially cooperative ties; (4) the replacement rule generates an implicit punishment that neutralizes exploitation and stabilizes cooperation. These insights have practical implications for designing policies or institutional mechanisms: simultaneous, population‑wide interventions are more likely to foster cooperation, whereas piecemeal, asynchronous changes may leave cooperation vulnerable to the initial conditions.