Index

The Prisoner's Dilemma

100%Open SVG ↗
THE PRISONER'S DILEMMAthe most famous game in the world — why two rational players both choose the outcome they both hate1 · THE STORYdetectiveyoupartnerarrested together, questioned in SEPARATE rooms.each can:• STAY SILENT — cooperate with your partner• CONFESS — defect, rat them out for a dealthe offer: rat while they stay silent → you walk FREE,they take the full 10 years. both rat → 5 each.both stay silent → just 1 year each (they can onlypin a minor charge). you can't talk to each other.2 · THE MATRIX (years in prison — LOWER is better)PARTNERSilentConfessYOUSilentConfess1 , 110 , 00 , 105 , 5NASH:both confessboth better offhere (1,1)!if partner is SILENT → confess (0) beats silent (1).if partner CONFESSES → confess (5) beats silent (10).so CONFESS is dominant for both → NE = (5,5).yet BOTH staying silent = (1,1) — better for everyone.rational self-interest → the outcome they both hate.3 · THE STRUCTURE — T > R > P > Srank the four outcomes (in DESIRABILITY):T Temptation — you defect, they cooperate (walk free)R Reward — both cooperate (1 year)P Punishment — both defect (5 years)S Sucker — you cooperate, they defect (10 years)any game with T > R > P > S is a Prisoner's Dilemma.defect strictly dominates → the unique NE is (P, P),even though (R, R) beats it for BOTH.individual rationality → COLLECTIVE irrationality.each doing the smart thing produces the dumb result.this is why the world has arms races and traffic jams.the ONE-SHOT dilemma is inescapable for rational players.the escape is REPETITION → board 04.4 · IT'S EVERYWHERE — the same trap, different costumesARMS RACEboth disarm = safe & cheap.but each fears the other arms,so both arm → broke & no safer.cooperate is best for all, defect is best for each.PRICE WARboth hold price = healthy margins.each undercuts to win share,both cut → margins gone.cooperate is best for all, defect is best for each.DOPING / CHEATINGclean sport = fair for all.each dopes to win,all dope → same ranks, worse health.cooperate is best for all, defect is best for each.CLIMATE / EMISSIONSall cut = livable planet.each free-rides on others' cuts,none cut → everyone loses.cooperate is best for all, defect is best for each.5 · TRAGEDY OF THE COMMONS — the n-player versionover-used →collapses foreveryonea shared resource (pasture, fishery, aquifer, shared DB) with open access:each user's rational move is to take a bit MORE (the gain is theirs, the cost isspread over all) → everyone over-uses → the resource is destroyed. Hardin, 1968.cousin: the FREE-RIDER problem — everyone benefits from a public good (clean air,open source, the on-call rota) but each prefers someone ELSE to pay for it.fixes: property rights / quotas, regulation, reputation, or making the resource excludable —i.e. CHANGE THE PAYOFFS so cooperating is each player's best move (board 07).PRISONER'S DILEMMAS IN YOUR STACK (shop / Agent)• SHARED INFRA = a commons: if every service disables its rate limit / retries aggressively, the cluster (the pasture) collapses — noisy-neighbour thrash. Quotas & backoff are the fences.• LOCAL OPTIMISATION: each team skips tests / adds a shortcut (individually rational under deadline) → global tech-debt rot. The 'defect' is cheap now and everyone pays later.• FREE-RIDING: everyone relies on the shared library / the one person who fixes flaky tests / the on-call hero — all benefit, few contribute. Make the contribution visible & rewarded (change payoffs).JARGON, DECODEDcooperate / defect — help the group / help only yourselfT > R > P > S — the payoff ranking that defines a PDdominant defection — defecting beats cooperating alwayscollective irrationality — smart individuals, dumb group resulttragedy of the commons — the n-player PD over a shared resourcefree-rider — benefits from a public good without payingpublic good — non-excludable, shared benefit (air, open source)collective-action problem — group can't achieve its shared interestone-shot game — played once; no future to enforce cooperation

Lessons that use this board

THE PRISONER'S DILEMMA the most famous game in the world — why two rational players both choose the outcome they both hate 1 · THE STORY detective you partner arrested together, questioned in SEPARATE rooms. each can: • STAY SILENT — cooperate with your partner • CONFESS — defect, rat them out for a deal the offer: rat while they stay silent → you walk FREE, they take the full 10 years. both rat → 5 each. both stay silent → just 1 year each (they can only pin a minor charge). you can't talk to each other. 2 · THE MATRIX (years in prison — LOWER is better) PARTNER Silent Confess YOU Silent Confess 1 , 1 10 , 0 0 , 10 5 , 5 NASH: both confess both better off here (1,1)! if partner is SILENT → confess (0) beats silent (1). if partner CONFESSES → confess (5) beats silent (10). so CONFESS is dominant for both → NE = (5,5). yet BOTH staying silent = (1,1) — better for everyone. rational self-interest → the outcome they both hate. 3 · THE STRUCTURE — T > R > P > S rank the four outcomes (in DESIRABILITY): T Temptation — you defect, they cooperate (walk free) R Reward — both cooperate (1 year) P Punishment — both defect (5 years) S Sucker — you cooperate, they defect (10 years) any game with T > R > P > S is a Prisoner's Dilemma. defect strictly dominates → the unique NE is (P, P), even though (R, R) beats it for BOTH. individual rationality → COLLECTIVE irrationality. each doing the smart thing produces the dumb result. this is why the world has arms races and traffic jams. the ONE-SHOT dilemma is inescapable for rational players. the escape is REPETITION → board 04. 4 · IT'S EVERYWHERE — the same trap, different costumes ARMS RACE both disarm = safe & cheap. but each fears the other arms, so both arm → broke & no safer. cooperate is best for all, defect is best for each. PRICE WAR both hold price = healthy margins. each undercuts to win share, both cut → margins gone. cooperate is best for all, defect is best for each. DOPING / CHEATING clean sport = fair for all. each dopes to win, all dope → same ranks, worse health. cooperate is best for all, defect is best for each. CLIMATE / EMISSIONS all cut = livable planet. each free-rides on others' cuts, none cut → everyone loses. cooperate is best for all, defect is best for each. 5 · TRAGEDY OF THE COMMONS — the n-player version over-used → collapses for everyone a shared resource (pasture, fishery, aquifer, shared DB) with open access: each user's rational move is to take a bit MORE (the gain is theirs, the cost is spread over all) → everyone over-uses → the resource is destroyed. Hardin, 1968. cousin: the FREE-RIDER problem — everyone benefits from a public good (clean air, open source, the on-call rota) but each prefers someone ELSE to pay for it. fixes: property rights / quotas, regulation, reputation, or making the resource excludable — i.e. CHANGE THE PAYOFFS so cooperating is each player's best move (board 07). PRISONER'S DILEMMAS IN YOUR STACK (shop / Agent) • SHARED INFRA = a commons: if every service disables its rate limit / retries aggressively, the cluster (the pasture) collapses — noisy-neighbour thrash. Quotas & backoff are the fences. • LOCAL OPTIMISATION: each team skips tests / adds a shortcut (individually rational under deadline) → global tech-debt rot. The 'defect' is cheap now and everyone pays later. • FREE-RIDING: everyone relies on the shared library / the one person who fixes flaky tests / the on-call hero — all benefit, few contribute. Make the contribution visible & rewarded (change payoffs). JARGON, DECODED cooperate / defect — help the group / help only yourself T > R > P > S — the payoff ranking that defines a PD dominant defection — defecting beats cooperating always collective irrationality — smart individuals, dumb group result tragedy of the commons — the n-player PD over a shared resource free-rider — benefits from a public good without paying public good — non-excludable, shared benefit (air, open source) collective-action problem — group can't achieve its shared interest one-shot game — played once; no future to enforce cooperation

Saved in this browser only.

← Dominant Strategies and Nash EquilibriumRepeated Games and Cooperation →