Knowing the past improves cooperation in the future

Cooperation is the cornerstone of human evolutionary success. Like no other species, we champion the sacrifice of personal benefits for the common good, and we work together to achieve what we are unable to achieve alone. Knowledge and information from past generations is thereby often instrumental in ensuring we keep cooperating rather than deteriorating to less productive ways of coexistence. Here we present a mathematical model based on evolutionary game theory that shows how using the past as the benchmark for evolutionary success, rather than just current performance, significantly improves cooperation in the future. Interestingly, the details of just how the past is taken into account play only second-order importance, whether it be a weighted average of past payoffs or just a single payoff value from the past. Cooperation is promoted because information from the past disables fast invasions of defectors, thus enhancing the long-term benefits of cooperative behavior.

💡 Research Summary

The paper investigates how incorporating information from past interactions into evolutionary game dynamics can promote cooperation in the future. Building on the classic two‑strategy social dilemma (cooperate vs. defect), the authors introduce two mathematically distinct but conceptually similar mechanisms that allow a player’s past payoffs to influence its current fitness, which in turn determines strategy updates.

The first mechanism uses a weighted moving average of the last M payoffs. A decay parameter α controls how quickly the weight of older rounds diminishes; α = 0 corresponds to the standard model with no memory, while α → 1 yields a very long memory window. The second mechanism selects a single past payoff Pτ instead of the current payoff with probability ν = exp(−τ/s), where s is set so that a payoff from 100 steps ago is chosen with 1 % probability. In both cases the past information is incorporated in a strategy‑neutral way—no explicit incentives or punishments are attached to any strategy.

Simulations are performed on a two‑dimensional lattice with periodic boundary conditions. Each player interacts with its four nearest neighbours, and the payoff matrix is fixed at R = 1 (mutual cooperation) and P = 0 (mutual defection). The remaining parameters S (sucker’s payoff) and T (temptation) are varied across the full range that defines the Prisoner’s Dilemma (T > R > P > S), Snowdrift (T > R > S > P) and Stag‑hunt (R > T > P > S) games.

The results, displayed in a series of (T,S) heat‑maps, show a clear monotonic increase in the stationary fraction of cooperators as the memory length grows (larger α) or as the time‑delay τ is set to an intermediate value. For the Prisoner’s Dilemma region—typically the most hostile to cooperation—cooperation can dominate the population when α ≈ 0.9 or τ ≈ 3, despite the high temptation to defect.

A microscopic analysis reveals why memory has this effect. The authors classify agents into four sub‑types: strong cooperators (C_C) whose current strategy matches their past strategy, weak cooperators (C_D) whose strategies differ, and analogously strong (D_D) and weak (D_C) defectors. When past payoffs are taken into account, the invasion probability of strong defectors into cooperative clusters is dramatically reduced, because the averaged fitness of a defector that has recently exploited cooperators is lowered by its poorer historical record. Conversely, weak cooperators act as a protective “shield” around strong cooperators; they may have suffered a temporary loss but still benefit from the long‑term payoff of being surrounded by other cooperators. Over time, weak cooperators become strong cooperators, allowing cooperative domains to expand slowly but steadily while defectors are unable to sustain their advantage. This asymmetric effect is evident in the transition rate plots (Fig. 5 and Fig. 6), where the net flow from defectors to cooperators becomes positive only when the memory parameter is sufficiently large.

Importantly, the two implementations—weighted average versus single past payoff—produce quantitatively similar outcomes, indicating that the precise mathematical form of memory is of secondary importance. What matters is the qualitative rule that past success is factored into present fitness assessments. This “strategy‑neutral” rule does not rely on additional incentives, punishment, or network heterogeneity, yet it robustly enhances cooperation across all three social‑dilemma types.

The discussion situates these findings within broader evolutionary theory, suggesting that cultural transmission, historical knowledge, and institutional memory in human societies may function analogously to the modeled memory mechanisms, stabilizing cooperative norms over generations. The authors propose future extensions such as varying network topologies, introducing more than two strategies, or testing the predictions in laboratory experiments with human subjects. Overall, the paper demonstrates that a simple incorporation of past payoff information can dramatically reshape evolutionary dynamics, slowing the spread of defectors and allowing cooperation to flourish in the long run.