Game Theory: Why Rational Choices Go Wrong

The argument started over a shared bowl of popcorn at a birthday party. Two children were told they could each take some now, or leave it and get a bigger refill later. Both reached in immediately. Both ended with less than they would have had by waiting, and both had done the sensible thing from where they were standing, because neither could trust the other to wait.

That is not a story about greedy children. It is a fully worked example of a real branch of mathematics, and the strange part is the conclusion. Sometimes the individually flawless decision produces a result that is worse for everybody than an option that was sitting right there. Game theory studies exactly those situations, and it is one of the few places in mathematics where the answer depends on what someone else is about to do.

The problem that gave the field its most famous example

In 1950, two researchers at the RAND Corporation built a small game to think about nuclear strategy. The version that survived was named by Albert Tucker, who framed it as two arrested accomplices held in separate rooms. Each is offered a deal: confess and testify, and you walk while your partner serves a long sentence. Stay silent, and if your partner also stays silent you both serve a short one.

The trap is visible once you lay out the four outcomes. Whatever the other person does, confessing gets you a shorter sentence. If they stay silent, confessing frees you instead of giving you a short term. If they confess, confessing is what saves you from the full sentence they are about to hand you. So a perfectly rational player confesses, and so does the other one, and both end up worse off than if neither had said a word.

The reference account of the prisoner's dilemma records the detail most retellings drop: John Nash, whose name is on the central concept of the field, pointed out that the rational answer in the repeated version of the game can be completely different from the rational answer in the single-round version. The one-shot trap is not the whole story.

What game theory is actually measuring

Strip away the criminals and the popcorn and there is one idea underneath, called a Nash equilibrium: a set of choices where no single player can do better by changing only their own move. Everyone is already doing the best they can given what everybody else is doing. The prisoner's dilemma is disturbing because its equilibrium is bad. Both confessing is stable, and both staying silent would be better, and no individual has a reason to move first.

The concept is older than Nash — the reference article on the equilibrium traces it to Cournot's 1838 work on competing firms. Nash's contribution was the guarantee. Every finite game, he proved in 1950 and 1951, has at least one equilibrium, provided you allow players to randomise — to pick a probability rather than a single move. That existence proof is why game theory became a tool rather than a collection of puzzles: you are promised an answer exists before you start looking.

Notice the shape this shares with a rule you may already know. A stable outcome where nobody benefits from moving alone is structurally like an equation that is satisfied — the same "this holds because nothing pushes it away" reasoning that appears in our systems of equations guide.

The experiment that changed the answer

For thirty years the dilemma stayed a warning about human selfishness. Then in 1980 a political scientist named Robert Axelrod ran a tournament. He invited researchers around the world to submit computer programs that would play the repeated game against each other, over and over, remembering what had happened before. Entries ranged from crude to elaborate, with sophisticated algorithms designed to detect and exploit opponents.

The winner was entered by Anatol Rapoport, and it was the simplest program in the entire contest — four lines of BASIC, by one count. It followed one rule: cooperate on the first move, then do whatever your opponent did last time. That is all. Against a cleverer algorithm, it lost. Over a long tournament against many strategies, it finished first.

Axelrod did not stop at the scoreboard. He analysed which features the successful strategies shared, and the conditions have a moral shape worth stating plainly. Be nice: never defect first. Be retaliating: if someone defects, respond so they notice. Be forgiving: once they cooperate again, let it go. Be clear enough that the other side can figure out what you are doing. Greedy strategies did badly, not because they were punished, but because the patient ones stopped cooperating and they ran out of partners.

The trick is the word “repeated”

Every phrase in those four conditions depends on one thing: the game is going to continue. In a single meeting, confession is correct and there is no counter-argument. In a relationship, a neighbourhood, or a class that meets again tomorrow, the arithmetic changes completely, because today's move becomes information about what tomorrow will be like.

The same structure explains a much larger version of the trap. When a resource is shared and each person benefits from taking a little more, the individually rational move is to take it, even though everybody taking is what destroys the resource. Ecologist Garrett Hardin gave this a name in a 1968 essay, and the account of the tragedy of the commons records that the idea stretches back to Aristotle. Overfishing, traffic, and a shared fridge at work all run on the same logic.

The reason this belongs in the same article is that its escape route is the one Axelrod found. Observers of the commons have argued for decades that collapse is not inevitable, because people who expect to keep meeting each other can work out restraint among themselves. The resource does not have to be policed by an outsider if the players understand that there is a next round. Repetition converts a one-shot grab into a standing relationship, and a standing relationship is what makes restraint pay.

That is the resolution of the children's popcorn problem, and it is not a lesson about being nice for its own sake. It is a statement about incentives. The reason to cooperate is that it makes future cooperation cheap and future defection expensive. Once you can be recognised and remembered, the rational move flips, and it flips because the game acquired a future rather than because anyone got kinder.

Where the reasoning breaks, and where it gets used anyway

The honest caveats matter, because this is a field that gets over-applied. The clean results assume players know the payoffs, that everyone is maximising the same kind of thing, and that the situation repeats. Change those and the conclusion weakens. Real people also weight fairness over gain, punish unfairness at their own cost, and misjudge what the other side wants, and none of that is captured by a payoff table. The mathematics of what should happen and the mathematics of what does happen are related but not identical, a distinction we have met before in the gap between theoretical and experimental probability.

The uses are everywhere anyway. Auctions use it, which is worth knowing before the next school fundraiser — the rules you choose decide how much people bid, and a badly designed auction collects less than a well-designed one for the same goods. Spectrum licences and government bond sales are designed with this mathematics, and the Nobel committee has handed out prizes for it repeatedly. In 1994 one of those prizes carried Nash's name, decades after a 1950 note that fit on a few pages.

It also explains something you may have felt without naming. Elections are a place where individually sensible ballots produce collective results nobody wanted, which is the same trap wearing a different costume — the territory our elections guide explores from the counting side.

Play one round tonight

You need no board. Give two children two cards each, one marked SHARE and one marked TAKE. Each secretly plays one card. If both SHARE, each gets two sweets. If both TAKE, each gets none. If one TAKES while the other SHARES, the taker gets three and the sharer gets none. Play it once and watch what happens, then play it five times in a row with the same partner and watch the difference.

That second round is where the lesson lands. Ask them afterward which version felt more like their real friendships, and whether they changed their card because they trusted the other player more or because they expected to meet them again. The answer is usually the second, and noticing it is the beginning of thinking strategically without thinking cynically.