Prisoner's dilemma explained: why cooperation is hard, and when repetition helps

What is the prisoner's dilemma?

In a one-round prisoner's dilemma, each player prefers defection regardless of what the other does, while both prefer mutual cooperation to mutual defection. The conflict is not simply that people are selfish. It is a specific ordering of incentives. A model explains something only if the actions and payoffs reasonably represent the situation you are studying.

The Stanford discussion of the prisoner's dilemma provides philosophical and game-theoretic context. Modic's prisoner's dilemma game uses three each for mutual cooperation, one each for mutual defection, and five versus zero when only one defects. Those are illustrative tokens. They make the conflict visible without pretending that trust, reputation, or human interests can always be measured by four numbers.

Why a personally tempting action can create a worse shared outcome

If your partner cooperates, defection earns five rather than three. If the partner defects, defection earns one rather than zero. In either case, defection gives the higher one-round token payoff. Yet when both follow that individual comparison, each receives one instead of the three each would receive under mutual cooperation. That is the precise tension in this particular matrix.

Before applying the story to a workplace, ask what the actions actually mean and whether this ordering holds. A disagreement might instead involve incompatible goals, uncertainty about quality, or a coordination problem where matching the other person's action is attractive. Calling all of those a dilemma can suggest the wrong remedy. Write the candidate matrix and inspect the alternatives before selecting the label.

A worked example: identical incentives, different joint results

Use a single round of the displayed token game. This is an invented payoff comparison, not a recommendation about how to treat a colleague. The purpose is to trace each player's reasoning explicitly rather than assume that a desirable shared result is automatically stable.

1. Suppose both cooperate: each earns three, so the joint total is six. Either player can switch alone and receive five while the other receives zero.
2. Suppose both defect: each earns one, for a joint total of two. Switching alone to cooperation would reduce the switcher's payoff to zero.
3. Compare the outcomes and the incentives separately. Mutual cooperation is jointly better, but the one-round incentive to depart from it remains under these stipulated payoffs.

What changes when the same players meet again?

In Modic's repeated game, compare strategies over a batch of rounds. Tit for tat starts by cooperating, then copies the opponent's previous actual move. That creates a connection between current conduct and later treatment. A move that looks attractive in isolation can affect the continuation of a relationship. This is a different decision environment from a single round with no further interaction.

Repetition does not guarantee cooperation. The future has to matter, actions need to be sufficiently observable, and participants may have different incentives or expectations. The Stanford discussion of repeated games and the folk theorem is useful context for why repeated interaction supports more than one possible result. A successful short run against one bot is not a proof of a universally best human strategy.

Add mistakes before trusting a simple reciprocity rule

Use the game's chance-of-flipped-move control. Intended cooperation can become actual defection, and the other strategy responds to what happened rather than what was meant. Reciprocal responses can then persist after a mistake. Compare the history and totals rather than reading one disappointing move as proof of a hostile intention.

This is why communication and error correction matter in real repeated relationships. A process for checking what happened can prevent a mistaken inference from becoming a sequence of retaliation. The model flips moves independently under a stated probability. It does not model apology, explanation, power imbalance, or strategic deception in a realistic organisation. Those omissions are reasons to ask better questions, not to treat the bot's policy as management advice.

Prisoner's dilemma versus stag hunt and the trust game

In the stag hunt game, the joint action becomes attractive when confidence that the partner joins is high enough. With joint reward eight and solo reward three, the joint action's expected payoff exceeds three when the partner's joining probability exceeds three-eighths. This is an assurance problem: the best response changes with the other player's likely action.

In the trust game, the sender can enlarge a shared pie by transferring resources but cannot force the recipient's return. The timing and distribution problem differ from choosing simultaneous cooperation or defection. These distinctions help diagnose whether a practical situation needs better assurance, a changed incentive, enforceable commitments, or information about what another person will do.

Map a cooperation problem before proposing a solution

Pick a recurring shared task and describe the available actions in concrete terms. Include what each person gains or loses, whether they meet again, what they can observe, and what mistakes are possible. If the apparent incentive comes from a reporting metric rather than the work itself, changing that metric may be more relevant than asking participants to become more cooperative.

Keep descriptions tentative. A person's preference for a safer action can reflect unreliable information or asymmetric consequences rather than an unwillingness to help. A useful model makes those possibilities inspectable. It should not be used to label a colleague a defector without checking whether the model's actions, payoffs, and assumptions actually fit their situation.

• What concrete actions correspond to cooperate and defect?
• Does defection really pay more under either partner action?
• Will these participants interact again, and how much does that matter?
• Can they distinguish a mistake from an intended action?
• Would changing assurance, information, or the payoff rule address the problem?

Learn from the games without confusing tokens with human motives

Run a small comparison at zero mistakes and record what you expect before playing. Then change only the error rate and compare the histories. Finally, move to the stag hunt and trust game to identify exactly which strategic feature changed. This sequence teaches more than collecting a high score without explaining why a strategy earned it.

The transferable skill is distinguishing a shared objective from the incentives sustaining it. When those are misaligned, goodwill may be insufficient; when they are already aligned, an adversarial story may be unnecessary. A careful explanation of a cooperation problem leaves room for both possibilities and for the practical institutions the simplified games do not include.