How Tit-for-Tat Won

Last lesson gave us the master key: repeat the prisoner’s dilemma and a defection today costs you all of tomorrow’s cooperation, so — when the future looms large enough — cooperating becomes the selfish move. But that only tells us cooperation can be rational. It doesn’t tell us which strategy actually does best when a crowd of them are thrown into a pit together. Is it the sly manipulator that probes for weakness? The paranoid fortress that never trusts anyone? The saint who forgives everything? For decades this was armchair speculation. Then, in the late 1970s, a political scientist did something almost nobody in game theory had done: he ran the experiment. The answer that came back was so simple, and so counterintuitive, that it reshaped how we think about cooperation across biology, economics, and diplomacy.

Axelrod ran a real tournament, not a proof

Here is the crucial thing to hold onto: what follows is an empirical discovery, not a theorem. Robert Axelrod, a political scientist at Michigan, wanted to know which strategy wins a repeated prisoner’s dilemma in practice. Rather than sit and prove it with algebra, he held a computer tournament — think of it as a chess championship where the players are little programs and the game is the iterated dilemma.

He wrote to leading game theorists, economists, mathematicians and computer scientists and asked each to submit a strategy encoded as a program: a rule that, handed the entire history of a match so far, outputs the next move — Cooperate or Defect. Fourteen entries came in, some of them fiendishly clever, built to probe an opponent, model its behaviour, and exploit any softness they detected.

Axelrod then ran a round-robin: every strategy played every other strategy (and a copy of itself, and a purely random player) in a long iterated dilemma of about 200 rounds. He used the standard payoffs — mutual cooperation pays 3 each, mutual defection 1 each, and a lone defector grabs 5 while the exploited cooperator gets 0. He summed each strategy’s total points across all its matches. Whoever accumulated the most points, over the whole gauntlet, won.

The winner was the shortest program submitted — a mere four lines of code — sent in by the mathematical psychologist Anatol Rapoport. It was called Tit-for-Tat.

Tit-for-Tat, and the four traits that won

The rule is almost embarrassingly small:

Cooperate on the first move. On every move after that, copy whatever your opponent did last time.

Be trusting to open; then simply mirror. That’s the entire strategy. Axelrod’s real contribution wasn’t the rule — Rapoport supplied that — it was dissecting why such a trivial program crushed the schemers. He found four properties, and every one of them is a transferable lesson for dealing with anyone you’ll meet again. Let’s take them one at a time, with a concrete micro-example each.

Nice — never defect first

A strategy is nice if it is never the first to defect. Tit-for-Tat opens with cooperation and never throws the first punch; the only way it ever defects is in response to your defection. This is what lets it strike up cooperation with anyone else willing to cooperate.

Micro-example. Tit-for-Tat meets Always Cooperate (a strategy that just cooperates no matter what). Round 1: both cooperate. Round 2: Tit-for-Tat copies its opponent’s last move (cooperate) → cooperates again. And so on forever: C-C, C-C, C-C… Both bank the reward of 3 every single round. Because Tit-for-Tat is nice, it never spoils a relationship that a cooperator was offering for free.

Retaliatory (provocable) — punish defection at once

A strategy is retaliatory (Axelrod’s word is provocable) if it punishes a defection immediately — the very next round — rather than turning the other cheek. This is what makes Tit-for-Tat impossible to exploit. Cheat it once and you eat a retaliation instantly, so cheating simply doesn’t pay.

Micro-example. Tit-for-Tat meets a sneaky exploiter — say, Always Defect, which defects every round hoping to feast on suckers. Round 1: Tit-for-Tat cooperates (nice), the exploiter defects and grabs 5 while Tit-for-Tat gets 0. But round 2: Tit-for-Tat copies that defection and defects too, so now they both get 1, and 1, and 1… The exploiter’s single early theft is the only profit it ever extracts. A pushover would have been robbed 200 times; Tit-for-Tat is robbed exactly once. That is the difference between nice and naive.

Forgiving — retaliate once, then let it go

A strategy is forgiving if, after punishing, it returns to cooperation the instant the opponent does — it doesn’t nurse the grudge. Tit-for-Tat retaliates for exactly as long as the opponent keeps defecting and not one round longer. The moment you cooperate again, it copies that and cooperates back.

Micro-example. Compare Tit-for-Tat with an unforgiving strategy, Grudger (a.k.a. Grim Trigger): Grudger cooperates until you defect once, then defects against you forever, no matter how sorry you are. Suppose an opponent defects a single time by accident and then offers cooperation again. Against Grudger, that one slip triggers permanent mutual defection — a feud with no exit, both stuck scoring 1 forever. Against Tit-for-Tat, the slip costs a single retaliatory round, then the offered cooperation is mirrored and both climb straight back to scoring 3. Forgiveness is what stops a stray defection from spiralling into an endless revenge loop — a property that becomes decisive in the next lesson, once mistakes enter the picture.

Clear — be legible so cooperation can be learned

A strategy is clear if it is simple and predictable enough that opponents can quickly figure it out and see that cooperation pays. Tit-for-Tat’s logic is transparent: hurt me and I hurt you back; be kind and I’m kind right back. Any opponent — even a crude learning program — reads that pattern within a few rounds and adjusts toward cooperating.

Micro-example. Contrast a clever, opaque prober that mixes in occasional sneaky defections to test you, disguised inside random-looking noise. Its opponents can’t tell whether a defection was a probe, a punishment, or a glitch, so they can’t learn how to earn its cooperation — and cooperation never stabilises. A punishment nobody can predict teaches nobody. Tit-for-Tat’s very dumbness is a feature: it’s a strategy you can read off in one glance, so the lesson “cooperate and I’ll cooperate” lands fast.

Here is the whole recipe in one table:

Play it — and watch the traits fire

Load the machine and run each experiment in the caption. You’ll literally see each of the four traits in action: niceness opening a relationship, retaliation slamming a cheat’s window shut, forgiveness rescuing a feud that an unforgiving Grudger would have let fester.

The counterintuitive result: nice strategies swept the board

Now the shock. When Axelrod tallied the final scoreboard, the entire top rank was occupied by nice strategies — every one of the eight highest scorers was a strategy that never defected first. All the clever, sneaky, exploitative programs — the ones expressly designed to probe and rob their opponents — finished in the bottom half. The meek didn’t just survive; they cleaned up.

Why on earth would not exploiting people be the winning move in a game that literally rewards exploitation? The answer is the deepest idea in the lesson: a nasty strategy poisons its own well.

Think about what a defector’s life is actually like across a whole tournament. Every time it meets a retaliatory strategy — and the field is full of them — its first-round theft triggers permanent punishment, and it spends the remaining ~199 rounds locked in mutual defection, both sides grinding out a miserable 1 point apiece. It does this against everyone. Its total is a long sum of ones. Meanwhile a nice strategy meets another nice strategy and the two of them settle into a long cooperative streak, harvesting 3 after 3 after 3 for 200 rounds. Nice strategies get rich off each other; nasty strategies condemn themselves to poverty with everyone. In a world containing enough cooperators, the cooperative harvest simply dwarfs whatever a cheat can steal.

The killer fact: Tit-for-Tat never won a single one of its head-to-head matches. Look closely and you’ll see it cannot: it’s never the first to defect, so the best it can ever do against any opponent is tie — and against a defector it loses that lone first round. So in every individual matchup, Tit-for-Tat either draws or comes second. And yet it won the whole tournament.

Let that sink in, because it’s the single most memorable thing in the theory of cooperation. Tit-for-Tat wins the war by never losing badly, not by ever beating anyone. The exploiters play to win each battle — and by winning battles they poison every relationship, so their total is small. Tit-for-Tat plays to build, not to beat: it draws its cooperative games at the high mutual-cooperation payoff and loses its rare bad games by a single round. Summed across the whole population, “never lose badly, cooperate wherever you can” beats “win every duel.” The lesson isn’t about being the toughest player in any given fight. It’s about which kind of player accumulates the most across a lifetime of encounters.

The second tournament — and the ecological rematch

A skeptic could object: maybe Tit-for-Tat only won because Axelrod’s particular fourteen entrants happened to suit it. So Axelrod did the honest thing — he ran it again, and this time much bigger. He published the first results, explained why Tit-for-Tat had won, and openly invited everyone to design something that would beat it. Sixty-two entries poured in from six countries, including many built specifically as Tit-for-Tat killers by people who now knew exactly what they were up against. Tit-for-Tat won again. Being nice, provocable, forgiving and clear survived even a field full of hunters gunning for it.

Then Axelrod ran the experiment that bridges us to the next lesson: the ecological tournament. Instead of one fixed round-robin, imagine the tournament played over and over across “generations,” where a strategy’s share of the population in the next generation grows in proportion to how well it scored in the current one. Strategies that do well become more common; strategies that do badly dwindle and vanish. This is a simple model of natural selection — and it turns the tournament from a one-off contest into an evolving ecosystem (exactly the population view we develop next lesson).

Watch what unfolds. Early on, the nasty exploiters do okay — there are still naive, over-nice strategies around for them to feast on. But every meal kills the food: as the exploiters drive their easy victims extinct, they run out of suckers, and now they mostly meet each other and other retaliators, dropping into mutual defection and scoring terribly. Their share collapses toward extinction. Meanwhile Tit-for-Tat, which was never anyone’s easy meal, keeps scoring well and keeps growing — and as the population fills up with fellow nice strategies, its cooperative streaks pay off even more richly. The nice strategies don’t just survive selection; they come to dominate the ecosystem, and the exploiters die out once their prey is gone.

Sort the behaviours by trait

Each behaviour below is a symptom of one of the four winning traits. Drop it into the right bucket. Guess before you check.

Match each trait to the disaster it prevents

The pitfall: winning a tournament is not being universally best

Here’s where we cash in the caveat from the top of the lesson. It is tempting — and wrong — to walk away believing “Tit-for-Tat is the optimal strategy, full stop.” It isn’t. It won those tournaments, against those fields. Two honest limits:

• It can’t invade some worlds when it’s rare. Drop a lone Tit-for-Tat into a population of pure defectors and it does worse than the natives — it eats an exploitation on the first round of every encounter and can never build the cooperative streaks that make it shine, because there’s no one to cooperate with. Tit-for-Tat’s strength is entirely relational: it thrives when there are other nice strategies to cooperate with, and withers when surrounded by defectors. Its success depends on the company it keeps. (How a cooperative cluster can nonetheless get a toehold against a sea of defectors is a whole mechanism we cover later — network reciprocity.)
• It is fragile to mistakes. Tit-for-Tat assumes it can perfectly observe what the opponent did. But the real world is noisy — signals get garbled, a cooperation gets misread as a defection, a finger slips. Because Tit-for-Tat retaliates for every perceived defection and never breaks a cycle unilaterally, a single mistaken defection between two Tit-for-Tat players sets off an endless, alternating revenge spiral — each punishing the other for the other’s last punishment, forever. The very forgiveness that saved it from a one-off feud isn’t quite enough once both sides can err. That crack is the entire subject of the next lesson.

When to use it

Reach for Axelrod’s four traits whenever you’re in a relationship you’ll return to — a supplier, a colleague, a neighbour, a rival, a country. Be nice (don’t defect first — you rarely gain what a soured relationship costs), be provocable (answer a genuine defection promptly, so you’re never anyone’s easy meal), be forgiving (drop the punishment the instant they come back, or you’ll feud your gains away), and be clear (make your policy legible so the other side can learn that cooperating with you pays). And remember the counterintuitive core: you’re not trying to beat the people you deal with — you’re trying to accumulate the most across all of them, which means never getting locked into a mutually destructive rut. Win the war by never losing badly.

But hold onto that second pitfall, because it’s about to bite. Tit-for-Tat’s whole logic assumes you can see clearly what the other side did. What happens when the world is noisy — when a cooperation gets misheard as a betrayal, when a hand slips? Next up, lesson 4: When Good Strategies Make Mistakes — where a single garbled signal traps two Tit-for-Tat players in an endless revenge spiral, and we meet the noise-proof refinements built to break out of it.