Skip to content
mentua.org

What Is the Prisoner's Dilemma?

Line drawing of a two-by-two grid in which arrows lead from a dotted circle in the top-left cell through two points to a filled disc in the bottom-right cellLine drawing of a two-by-two grid in which arrows lead from a dotted circle in the top-left cell through two points to a filled disc in the bottom-right cell

6 min read≈ 1,220 words3 questions · up to 3 points

What this is about

Two players who both think carefully can end up with a result that both of them like less than another one that was open to them. You learn the arithmetic behind this with one small table of points, why a computer tournament around 1980 was won by the simplest program, and why real people are less trapped than the model says.

In a nutshell· skip ahead

The prisoner's dilemma is a game in which each of two players does better by betraying the other, whatever the other does. So both betray, and both end up worse off than if both had cooperated. When the same players meet again and again, cooperation can pay, because each betrayal can be answered in the next round.

Two suspects are questioned in separate rooms. Each is told: if you testify against your partner while your partner stays silent, you go free and your partner gets a long sentence. If both stay silent, both get a short sentence, and if both testify, both get a medium one. Each suspect thinks carefully, and both testify. Why do two careful thinkers choose a result that is worse for both of them than staying silent?

Why do two sensible players end up worse off than they had to?

This situation is called the prisoner’s dilemma: a game in which each of two players can either cooperate or betray the other, which game theorists call “defecting”. For each player, defecting pays more whatever the other one does, so both defect and both end up worse off than if both had cooperated.

We can see this with points instead of prison years, using the scores that the political scientist Robert Axelrod used in his famous tournaments:

Ben cooperates Ben defects
Anna cooperates Anna 3, Ben 3 Anna 0, Ben 5
Anna defects Anna 5, Ben 0 Anna 1, Ben 1

Now let us follow Anna’s reasoning.

  1. Suppose Ben cooperates. Anna gets 3 if she cooperates too and 5 if she defects, so she should defect.
  2. Suppose Ben defects. Anna gets 0 if she cooperates and 1 if she defects, so again she should defect.
  3. Defecting is better in both cases. A choice that is best whatever the other player does is called a dominant strategy.
  4. Ben reasons the same way, because the table is symmetric.
  5. Both defect and get 1 point each, although both cooperating would have given them 3 each.

The result in step 5 is a Nash equilibrium, named after the mathematician John Nash: a pair of choices where neither player can do better by changing only their own choice. If Anna switched to cooperating while Ben defects, she would drop from 1 to 0. The equilibrium is stable, but it is bad for both, and that gap between what is stable and what is best is the dilemma.

But what if Anna and Ben meet again tomorrow, and again the day after? In a repeated game, a defection today can be answered tomorrow. Axelrod asked game theorists to send in computer programs to play each other about 200 rounds at a time, and the winner was Tit for Tat, sent in by the game theorist Anatol Rapoport. It cooperates in the first round and then copies whatever the other player did in the round before.

Over 200 rounds, two Tit for Tat players cooperate every time and earn 600 points each. Against a program that always defects, Tit for Tat loses the first round (0 against 5) and then defects back, so it ends with 199 points against 204. Two programs that always defect earn only 200 each. Tit for Tat never scores more than the program it plays against, yet it earned the most in total, because it collected the high points with every program that was willing to cooperate. Axelrod ran a second tournament with 62 entries, all of them knowing the first result, and Tit for Tat won again.

It succeeds by eliciting cooperation from others, not by defeating them.

Robert Axelrod, “The Evolution of Cooperation”, adapted from his 1984 book

Cooperation in the repeated game has one condition, because if both players know exactly which round is the last, each defects in that round, then in the round before, and cooperation comes apart back to the first move. Cooperation needs what he called a large “shadow of the future”: a good chance that the same players meet again, and a next meeting that matters enough to them.

What the prisoner’s dilemma teaches you

The prisoner’s dilemma teaches you to ask two questions when people fail to cooperate: does each side gain by defecting whatever the other does, and will they meet again? If the first answer is yes and the second is no, you should expect defection even among decent people.

  • Trade and arms. Axelrod’s own example is two countries that would both gain from free trade, although each is better off keeping its trade barriers whatever the other does. Arms races follow the same pattern.
  • The trenches of World War I. Axelrod describes how small units on the Western Front often avoided shooting to kill as long as the other side held back too, because the same units faced each other for months and every attack could be answered.
  • Your own work. With a colleague you will work with for years, Tit for Tat is a sensible guide: start by cooperating, respond quickly when someone takes advantage of you, forgive once they change, and be easy to read. In a one-time deal with a stranger, the protection must come from a contract or a deposit instead.

The same logic appears in biology. Help between relatives can pay through shared genes, as Kin Selection and Hamilton’s Rule, Explained shows. Help between unrelated animals needs the repeated game instead.

Do real people play the way the model predicts?

They do not, and this is the first limit. David Sally reviewed laboratory experiments with prisoner’s dilemmas published between 1958 and 1992 and concluded that a model of pure self-interest usually fails to match what people do. The gap was widest where players could talk to each other before they chose.

The second limit concerns the model itself. Its rules are fixed: the players cannot talk, cannot make agreements and cannot change the payoffs. The political scientist Elinor Ostrom argued that many real problems look like a prisoner’s dilemma only from outside. The fishers and farmers she studied talked, agreed on rules and punished those who broke them, so they changed the game instead of being trapped in it.

A common misunderstanding is that the prisoner’s dilemma proves people are selfish or that cooperation is irrational. It proves neither, because it only says what follows if players count nothing but their own points, meet once and cannot make promises. A last limit concerns Tit for Tat: if moves are sometimes misread, a single mistake can lock two Tit for Tat players into an endless chain of revenge, so more forgiving strategies do better when errors happen.

Learn more

The game has no single inventor. Merrill Flood and Melvin Dresher, two mathematicians at the RAND Corporation in California, devised it in early 1950 and had colleagues play it one hundred times in a row. In the same year, Albert Tucker, a mathematician at Princeton, told it as a story about two prisoners for an audience of psychologists at Stanford. The story gave the game its name, which first appeared in print in 1957. Tucker’s doctoral student John Nash defined his equilibrium in a dissertation written that same year.

The same trap with many players is described in What Is the Tragedy of the Commons?, and for shared goods that nobody wants to pay for in What Is the Free-Rider Problem?. For how real communities escape it, read “Governing the Commons” by Elinor Ostrom: Summary and Review, which treats the prisoner’s dilemma as one of three models that Ostrom wants to correct. Axelrod’s short book from 1984 is the classic next step, reviewed in “The Evolution of Cooperation” by Robert Axelrod: Summary and Review.

Quiz

3 questions · up to 3 points

Open a question and pick one answer. Only your first answer counts: +1 if it is right, −1 if it is wrong.

  1. In a one-round prisoner's dilemma, why do both players end up defecting?
    Show answer

    Whatever the other player does, defecting gives each player more points than cooperating. Defecting is a dominant strategy. Against a cooperator it gives 5 instead of 3, and against a defector 1 instead of 0.

  2. Two petrol stations stand on the same road. If both keep prices high, both earn well, but each can win customers by cutting its price. What does the prisoner's dilemma suggest?
    Show answer

    If they meet only once, both cut prices, but if they compete for years, both may keep prices high by answering every cut. This is the one-round game and the repeated game in one case. A long future makes a price cut answerable, so cooperation can last.

  3. How did Tit for Tat win Axelrod's computer tournaments?
    Show answer

    It got other strategies to cooperate with it, although it never scored more than the strategy it was playing against. Tit for Tat can at best tie in any single match. It won on its total because it earned the cooperation payoff with every friendly strategy.

The takeaway

The prisoner's dilemma is a game in which betraying is each player's best move whatever the other does, so two sensible players both betray and get less than if both had cooperated. With 3 points each for cooperating, 1 each for betraying and 5 against 0 when only one betrays, betraying always pays more in a single round. In repeated play, Tit for Tat won Axelrod's tournaments by cooperating first and copying the other's last move. Real people cooperate more than the model predicts, and many real situations let them talk and change the rules.

References

  1. Prisoner's Dilemma. Steven Kuhn, Stanford Encyclopedia of Philosophy

    A thorough academic overview: the history at RAND, Tucker's story, the formal conditions, repeated play and Axelrod's tournaments.

  2. The Evolution of Cooperation. Robert Axelrod, Essay adapted from his 1984 book, in Breakthrough: Emerging New Thinking (Stanford University)

    Axelrod's own short account of the tournaments, the four rules of successful strategies and the truce in the World War I trenches.

  3. Conversation and Cooperation in Social Dilemmas: A Meta-Analysis of Experiments from 1958 to 1992. David Sally, Rationality and Society 7 (1), 58–92, 1995

    The abstract of a review of 35 years of laboratory experiments, which finds that pure self-interest rarely predicts what people do.

prisoners-dilemmagame-theorynash-equilibriumdominant-strategycooperationdefectiontit-for-tatrepeated-gamesrobert-axelrodanatol-rapoportalbert-tuckerrand-corporationevolution-of-cooperationsocial-dilemmasreciprocityelinor-ostromcollective-action