Two suspects are held in separate rooms. The police can convict both of a minor charge, but they want a confession for something bigger, so they offer each of them the same deal.
Give evidence against the other and you walk free while they take the full sentence. If you both confess, you both go down, with a little time off. If you both stay silent, you both serve the short sentence for the minor charge.
Scored from best to worst for you: walking free is 3, the short sentence is 2, both confessing is 1, and taking the full sentence while the other walks is 0.
You are Suspect A
Watch one round, or choose now.
Suspect A
or
You
Suspect A
?
You
Suspect B
or
The other suspect
Suspect B
?
The other suspect
The other suspect picks at random here, like a coin toss.
The whole game on one grid
Every pair of choices and what each side gets: yours first, then theirs. Each number is a payoff.
Step 0 of 3
The cards are face down. Turn them, then step through which squares are stable.
What do you think they'll do?
Your best move depends on what you expect from the other suspect. Drag to set how likely you think it is they pick Stay silent.
What it teaches
Individually rational choices can make everyone worse off, and talking about it does not help.
The usual mistakeLabelling any situation with a bad shared outcome a prisoner's dilemma, including stag hunts, coordination failures and arms races where no player has a dominant strategy, and so prescribing enforcement where reassurance was what was needed.
The checkCheck for a dominant strategy before using the label. If what you should do depends on what the other does, it is not this game and the remedy is different.
What people actually do
The name of the game changes how it is played
The theory says
A game is defined by its players, actions and payoffs. Two presentations with identical payoff matrices are the same game, so labels, descriptions and irrelevant context should not change behaviour.
People actually
A one-shot prisoner's dilemma labelled the Community Game draws roughly twice the cooperation of the same game labelled the Wall Street Game, and labels predict individual behaviour better than observers' judgements of who is cooperative. In paired treatments where a payoff parameter changes without moving the equilibrium, behaviour tracks the parameter rather than the prediction, as with claims in the travellers' dilemma moving with the size of the penalty.
Why is still argued over.
People cooperate in games where defection strictly dominates
The theory says
In a one-shot prisoner's dilemma defection strictly dominates, so no belief about the opponent can make cooperation a best reply and the unique equilibrium is mutual defection. In a finitely repeated version with a known end, backward induction gives defection at every round.
People actually
Cooperation rates in one-shot play are typically between a quarter and a half, and in finitely repeated play cooperation is high early and collapses only in the final rounds, so the unravelling appears at the end of the sequence rather than throughout it.
Why is still argued over.
In the real world
Real world
Doping in professional cycling
Doping was systematic in professional road cycling through the 1990s and 2000s, established by the Festina affair, the Operacion Puerto investigation and the United States Anti-Doping Agency's case against Lance Armstrong, who was stripped of his Tour de France titles in 2012. Riders who have since described the era talk about a choice between doping and losing their place in the sport rather than a choice about whether to win, and the revealed preference of the peloton looks like keeping a contract rather than winning a race. Testing, the biological passport and the threat of bans are all attempts to change the payoffs from outside the race, because nothing inside it could.
Doping pays whatever the rival does, since it either wins the race or merely keeps you level, so the temptation is a dominant strategy rather than the mutual-confidence problem of a stag hunt where the costly action only pays if the other takes it too.
Real world
Investing in Freedonia
Player 1
the foreign investor
Player 2
the host government
History
An arms race
Player 1
State A
Player 2
State B
Cooperate
disarm
Defect
arm
Fiction: The Dark Forest
The dark forest
Two axioms, that survival comes first and that matter is finite, are combined with what the novels call the chain of suspicion: neither civilisation can establish the other's intentions, and neither can rule out that the other will outgrow it faster than light can carry a message. The conclusion drawn in the books is that any civilisation which reveals its position should expect to be destroyed without warning, and the later events of the trilogy bear that out. The reasoning is presented as a derivation rather than a disposition, which is the point of it.
The novels' argument is that striking is better whatever the other does, because a benign neighbour costs nothing to remove and a hostile one must be removed, so it is a dominant strategy rather than the failure of mutual confidence that would make it a stag hunt.
Braess's paradoxThe inefficiency is created by adding an option rather than by the structure of two existing ones, so the remedy is to take something away rather than to enforce something. Not on the wall yet.
Cheap talkThe players hold different information rather than the same information, which is why talk can do something here and does nothing there. Not on the wall yet.
DeadlockThe dominance equilibrium is Pareto efficient, so the outcome is not a tragedy. Not on the wall yet.