Axelrod's first Prisoner's Dilemma computer tournament turned a repeated strategic game into a systematic contest among programmed rules of action. Its striking result was that Tit for Tat, a strategy that cooperates first and then copies the opponent’s previous move, won overall; more broadly, all eight strategies that never defected first occupied the tournament’s top eight places. The contest made cooperation’s success measurable not as a moral claim, but as an outcome shaped by repeated interaction, retaliation, and the possibility of continuing play.

From a simple game to a computer contest

The tournament grew from the Iterated Prisoner’s Dilemma, whose simpler, one-off form gives each player only two options: cooperate or defect. In the original human version, a banker adjudicates the result. Each player places one of two cards face down, so the choices are made simultaneously and neither player knows the other’s move until the banker turns over the cards. The suspense comes from the fact that each player’s winnings depend on both choices, not just their own.

Repeating the game changes the strategic problem. Players can respond to the history of earlier moves, so strategies need not be limited to unconditional cooperation or defection. They can follow rules that copy an opponent, remember an earlier betrayal, or introduce an occasional defection. The possible rules are limited chiefly by human ingenuity, and it is not self-evident which one will perform best over many different encounters.

Axelrod set out to compare such rules by inviting experts in game theory to submit strategies. These were not merely descriptions of how a person might behave: each was a preprogrammed rule for action, submitted in computer language. Fourteen strategies arrived. Axelrod added a fifteenth, Random, which chose between cooperation and defection at random and served as a baseline against which the designed strategies could be judged.

How the tournament measured success

Axelrod translated all fifteen entries into a common programming language and ran them against one another on a single computer. Every strategy played each of the others, including a copy of itself. With fifteen strategies, this produced 225 games. Each pairing lasted 200 moves; after all the games were complete, the computer totalled the winnings.

The contest ranked strategies by their cumulative points across all fifteen pairings, rather than by whether they could exploit any single opponent. The scoring scheme awarded three points for mutual cooperation, five for the Temptation to defect, one for mutual defection, and zero for the Sucker’s payoff. Thus a one-move gain from exploiting a cooperative opponent could be outweighed by the poor returns of later mutual recrimination.

A strategy could theoretically score as much as 15,000 points: 200 moves at five points per move against each of fifteen opponents. The practical benchmark was much lower. Two players who cooperated consistently would each earn 600 points over a 200-move game, or three points per move. No strategy’s average score exceeded that benchmark. The tournament therefore rewarded performance across a mixed field of opponents, not the theoretical maximum available from uninterrupted exploitation.

Tit for Tat and the value of a first move

The winning entry was Tit for Tat, submitted by Professor Anatol Rapoport, a psychologist and games theorist from Toronto. Its rule was strikingly spare: cooperate on the first move, then copy the other player’s previous move. It does not begin by attacking, but it is capable of retaliation; its response is determined by what the opponent has just done.

Against a copy of itself, Tit for Tat produced an unbroken sequence of cooperation. Both players cooperated on the first move, and on every subsequent move each copied the other’s cooperation. Each finished with 600 points, the full 100 per cent benchmark. The self-play result matters because each strategy in the tournament was tested not just against competitors but against its own kind: a rule able to sustain cooperation with a like-minded counterpart could earn the reward throughout the game.

A hypothetical encounter with Naive Prober shows how a single unprovoked defection could disrupt that return. Naive Prober resembles Tit for Tat but defects at random from time to time, perhaps once in ten moves. Until it defects, the two strategies cooperate. If Naive Prober defects on the eighth move, Tit for Tat cooperates and receives zero, while Naive Prober takes the five-point Temptation payoff. On the next move, Tit for Tat copies that defection; Naive Prober copies Tit for Tat’s earlier cooperation, leaving Naive Prober with zero and Tit for Tat with five. The alternating retaliation that follows gives each an average of 2.5 points per move, below the three points they could have maintained through mutual cooperation.

The tournament’s pattern: nice, but not endlessly forgiving

Axelrod classified entries by their general properties rather than treating each program’s coding as the main point. The most important category was “nice”: a nice strategy never defects first. Tit for Tat qualifies, because it defects only in retaliation. Naive Prober and Remorseful Prober count as “nasty” because each sometimes defects without having been provoked, however rarely.

The rankings sharply divided these two groups. Eight of the fifteen entries were nice, and those same eight were the tournament’s top-scoring strategies; the seven nasty strategies came in well behind them. Tit for Tat scored an average of 504.5 points, or 84 per cent of the 600-point benchmark. The other nice strategies ranged from 83.4 down to 78.6 per cent, while Graaskamp, the most successful nasty strategy, scored 66.8 per cent. The outcome was not that every cooperative rule performed identically, but that the field’s better performers shared a refusal to initiate defection.

Being nice was not the same as remembering every offence forever. Axelrod also distinguished strategies by whether they were forgiving: a forgiving strategy may retaliate, but has a short memory and can resume cooperation. Tit for Tat answers a defection immediately, then lets the next move reset the relationship if the opponent cooperates. Grudger, a strategy discussed earlier in the book, never forgets a defection. A formally identical tournament entry, called Friedman, was technically nice because it never defected first, yet it did poorly among the nice strategies. Once Grudger or Friedman enters mutual recrimination with an opponent, neither can readily restore cooperation, even if the opponent becomes remorseful.

Tit for Two Tats represents an even more forgiving rule: it tolerates two consecutive defections before retaliating. Although no contestant submitted it, Axelrod calculated that it would have won the tournament, because its added tolerance would have helped prevent long runs of mutual recrimination. Forgiveness here is not simply leniency as a virtue; it is a strategic capacity to escape a destructive sequence and return to the better average offered by cooperation.

What the result shows about cooperation

The tournament’s result depends on iteration. If players know exactly when a game will end, the final move becomes equivalent to a one-off Prisoner’s Dilemma, and backward reasoning makes defection the rational move on that last turn. The same reasoning then reaches the penultimate turn and continues backward through the game. Repeated cooperation therefore requires that players not know precisely when their interaction will end; the future must remain an uncertain possibility rather than a countable series of moves.

This condition gives meaning to the “shadow of the future”: the prospect of further encounters allows a present response to affect later outcomes. Tit for Tat’s first cooperative move creates an opening for mutual gain; its immediate retaliation makes an unprovoked defection costly; its willingness to copy cooperation again prevents retaliation from becoming permanent. The tournament thus linked its numerical rankings to a particular structure of interaction, in which a strategy can be both responsive and capable of returning to cooperation.

The book applies this logic beyond computer programs and human players. It argues that human disputes can sometimes be transformed from apparent zero-sum confrontations into nonzero-sum arrangements. Divorce, for example, can remain cooperative in some respects even when a marriage has broken down, including for the sake of children and family finances. Yet in England and, until recently, all fifty states of the United States, professional rules required a lawyer to accept only one member of a couple as a client. The arrangement could divide the couple into opposing legal sides, even when they wanted an amicable settlement; the book notes that some lawyers nevertheless encourage clients to settle out of court.

The same contrast appears in the account of the 1977 English Football League season. On 18 May, Bristol and Coventry were playing a crucial match while Sunderland played elsewhere. If Sunderland lost, Bristol and Coventry both needed only a draw to avoid relegation. Their match was five minutes behind schedule, so the result from the other ground arrived before the Bristol-Coventry game ended. With two minutes remaining, the news that Sunderland had lost was displayed on the stadium’s electronic message board. The incentives changed: scoring was no longer necessary, and trying for a goal could expose either team to the risk of defeat and relegation. Players who had been competing fiercely instead sought to preserve the draw, while supporters who had been rivals joined in a shared celebration. A contest that had appeared zero-sum had become one in which both sides could benefit from the same outcome.

These examples illuminate the scope and the limit of the tournament’s lesson. A strategy like Tit for Tat cannot guarantee cooperation in every setting; it performs within a repeated game whose participants can respond to one another and whose ending is not known in advance. But where those conditions hold, rivals need not benefit only by injuring one another. Axelrod’s computer contest made that possibility legible in points and rankings: across 225 games, the winner was a rule that began cooperatively, answered defection, and could continue cooperating when the other player did.