Skip to main content

Repeated Prisoner's Dilemma: From Myopia to Reciprocity

When the Prisoner's Dilemma is repeated, strategy is no longer a single move. It becomes a policy over time: "what to do now given what happened before."

What Changes When the Game Is Repeated​

  • Future consequences matter ("shadow of the future")
  • Reputation and retaliation become possible
  • Conditional cooperation strategies can emerge (for example, reciprocity)

Finite vs Ongoing Interaction​

  • Finite known horizon: backward-induction logic pushes toward end-game defection, often cascading earlier.
  • Uncertain or effectively ongoing horizon: cooperation can be sustained if future interaction is valuable enough.

This tension is exactly why repeated social dilemmas are useful for studying learned cooperation.

Why MARL Is the Bridge​

In practice, agents do not compute perfect equilibria analytically. They learn from interaction.

Multi-Agent Reinforcement Learning (MARL) lets us test whether cooperative conventions can emerge from repeated play under bounded learning dynamics, exploration, and non-stationarity.

Next step: PPO Study