In 1980, Robert Axelrod invited game theorists to submit computer programs for a tournament. The game was the iterated prisoner's dilemma. Each program would play against every other program and against itself, 200 rounds per match. The program that accumulated the most points across all matches would win.
The winning program was four lines long. It cooperated on the first move, then on every subsequent move did whatever its opponent had done on the previous move. That's the whole strategy. Cooperate first; after that, mirror. Anatol Rapoport called it Tit-for-Tat.
The paradox: Tit-for-Tat cannot outscore its opponent in any single match. If the opponent always cooperates, they tie — both cooperating every round, both earning 3 points per round. If the opponent always defects, Tit-for-Tat gets exploited once and then defects in response — the opponent always ends up with one more point than Tit-for-Tat (the opening round where Tit-for-Tat was nice). Against any opponent, Tit-for-Tat always mirrors rather than leads, so it can tie or lose individual matches but it cannot win them. It can never score more than its opponent.
Yet it won the tournament. More precisely: it accumulated more total points across all its matches than any other program.
The prisoner's dilemma is a payoff structure. Two players simultaneously choose to cooperate or defect. If both cooperate: both get a moderate reward (3). If both defect: both get a low reward (1). If one defects and one cooperates: the defector gets a high reward (5), the cooperator gets nothing (0). The structure makes defection individually rational — regardless of what your opponent does, you're better off defecting. But if both reason this way, both get 1. If both had cooperated, both would have gotten 3. The collectively optimal outcome is individually irrational.
In a single encounter, defection dominates. Axelrod's tournament was iterated — 200 rounds, same partner — which changes the structure. Your opponent will remember what you did. Defection can be punished. Cooperation can be rewarded. The shadow of the future makes cooperation possible.
The fourteen programs submitted included sophisticated strategies with long memory, probabilistic moves, attempts to detect and exploit cooperators. Tit-for-Tat beat all of them. Axelrod ran a second tournament after publishing the results, with 62 entrants who knew Tit-for-Tat had won. They designed their strategies specifically to do better than it. Tit-for-Tat won again.
Axelrod identified four properties of Tit-for-Tat that he thought explained its success. Nice: it was never the first to defect. Forgiving: after punishing a defection, it would immediately return to cooperation if the opponent cooperated — no extended grudges. Retaliatory: it punished defection immediately, not after several rounds of being nice. Clear: its behavior was simple enough that opponents could figure out what it was doing and respond accordingly.
The nice property eliminated the possibility of mutually destructive defection spirals from the start. The forgiving property allowed relationships to recover after defections rather than locking into permanent punishment cycles. The retaliatory property meant that exploitation was reliably costly to the exploiter. The clear property meant that a rational opponent would recognize Tit-for-Tat's incentive structure and cooperate — there was nothing to gain by defecting.
Most of the complex strategies failed because they tried to do too much. Some attempted to detect exploitable cooperators and take advantage of them — but Tit-for-Tat wasn't exploitable, so this cleverness backfired when these strategies faced each other or faced Tit-for-Tat. Some were unforgiving — punishing defection permanently — and locked into mutual defection with anyone who had ever defected once. Friedman's GRIM strategy, which defected forever after any defection, had the highest possible retaliation but paid for it with destroyed relationships.
The strategies that did worst were not the dumbest. Some were quite sophisticated. They lost because they tried to win each encounter, and that's not what the tournament was for.
The paradox doesn't disappear when you explain it. Tit-for-Tat wins a tournament in which it cannot win a single match. Its advantage is not in any particular encounter but in the aggregate — in what kinds of relationships it tends to establish. Against cooperators, it maintains cooperation and everyone earns well. Against defectors, it stops being exploited after one round and moves on. Against strategies that are occasionally defective, it punishes and forgives, finding equilibrium.
A strategy that always defects beats Tit-for-Tat in their individual match. But in a world where you play against many different opponents and keep meeting them, always-defect does badly: it earns 5 from the cooperators in the first round and then earns 1 in every subsequent round as they retaliate. Meanwhile, the cooperating strategies that found each other are earning 3 per round, session after session. The defectors' advantage is front-loaded and self-limiting. The cooperators' advantage compounds.
What Axelrod's tournament showed was not that cooperation is more virtuous than defection — that would be a moral claim, not a game-theoretic one. What it showed is that under conditions of repeated interaction with the same set of agents, the strategies that enable cooperation outperform the strategies that exploit it, even on a purely payoff-maximizing criterion. No external enforcement needed. No altruism required. The structure of the game itself rewards trustworthiness over time.
This is why the result mattered for biology. Reciprocal altruism — I help you now, you help me later — had been a puzzle since Trivers proposed it in 1971. How could natural selection preserve a behavior that is costly in the moment? The iterated prisoner's dilemma gives an answer: if organisms interact repeatedly and can recognize each other, cooperative strategies can be stable against invasion by defectors. Mutualism, alarm calls, blood sharing in vampire bats — all can in principle be sustained by the same logic Axelrod demonstrated with computer programs playing a tournament.
The simplest strategy in the tournament was the best one. That's not always true — in more complex games, more complex strategies can outperform. But in the first tournament, where nobody knew what kind of strategies would be submitted, the simplest possible conditional strategy beat every attempt at cleverness.
Tit-for-Tat never scored more than its opponent. It won by making cooperation cheap and defection costly, never first.