
what game theory taught me about people
A reflection on the Prisoner's Dilemma, cooperation, and why the nicest strategy often wins — without being naive.
In 1980, a political scientist named Robert Axelrod ran a tournament.
He invited economists, psychologists, mathematicians, and game theorists to submit computer programs. Each program would play a simple game against every other program, repeatedly. No communication, no explanation, no appeal. Just choices.
The game was the Prisoner’s Dilemma.
You and another player each choose, simultaneously: cooperate or defect. If you both cooperate, you both get a moderate reward. If you both defect, you both get almost nothing. But if one of you defects while the other cooperates, the defector wins big and the cooperator gets nothing.
The temptation is always to defect. Whatever the other person does, defecting is the better move in that single round. If they cooperate and you defect, you win more. If they defect and you defect, you avoid the worst outcome.
The logic is airtight. And yet.
The winning strategy, submitted by the mathematician Anatol Rapoport, was fourteen lines long.
It was called Tit for Tat.
It cooperated on the first move. Then it did whatever the other player did on the previous move. That was the entire strategy. No complex predictions. No elaborate models of the opponent. Just: start nice, then mirror.
It beat every other entry in the tournament. Then Axelrod ran a second tournament, told everyone Tit for Tat had won the first one, and invited new strategies designed specifically to beat it.
Tit for Tat won again.
When Axelrod analysed why, four properties emerged from the strategies that did well. They were not what most people would have predicted.
Be nice. Never defect first. The best strategies did not start by trying to exploit anyone. They opened with cooperation and waited to see what would come back.
Be provocable. When someone defects against you, respond immediately. Do not absorb betrayal quietly. Do not keep cooperating in the hope they will eventually come around. Retaliate once, clearly, so the other player knows the behaviour has a consequence.
Be forgiving. One retaliation is enough. After you have signalled that defection has a cost, return to cooperation. Do not carry a grudge indefinitely. Sustained punishment punishes you as much as it punishes them — you both lose rounds you could have spent cooperating.
Be clear. The most successful strategies were simple enough that the other player could figure out what you were doing. If your strategy is too clever, the opponent cannot recognise it, cannot respond to it, and cannot cooperate with you even if they want to.
What is interesting about this list is what is not on it. Nowhere does it say: be aggressive. Be suspicious. Maximise every advantage. Get yours before they get theirs.
The mathematics, run across thousands of iterations, produced a strategy that would not look out of place in a guide to getting along with people.
The practical translation of these four properties changes depending on the situation, but the structure holds.
In a negotiation, being nice means opening with a reasonable position — not the most aggressive anchor you can justify. Being provocable means that when the other side makes an unreasonable move, you do not absorb it silently: you respond in kind, once, so the signal is clear. Being forgiving means that when they return to a reasonable position, you do too — you do not punish them for the earlier move after they have already adjusted. Being clear means your counteroffers are predictable enough that they can model what a fair outcome looks like.
In a working relationship, the payoff matrix changes but the structure does not. If someone misses a commitment, extracts more than they contribute, or shifts the terms after they were agreed — the correct move is one proportionate response. Not silence. Not an escalating campaign. One clear signal. Then return to the default.
What the strategy is doing in both cases is communicating a payoff structure. Cooperate, and this is what you get. Defect, and this is what you get. When that structure is transparent and consistent, the other party can calculate the long-run expected value of cooperation and compare it against the one-time gain from defection. Compounded across repeated interactions, cooperation almost always wins.
The instinct to harden early — to stop extending trust before it has been earned, to open defensively — forecloses the cooperative equilibrium before the other party has had a chance to find it. You end up in a mutual defection state that neither side actually wanted.
Most actors will choose cooperation when the payoff structure is clear. The strategy is optimised for that majority. It is not designed to survive indefinitely against someone who will defect regardless.
The harder question is what to do in an environment that is genuinely hostile. Where the people around you are not occasionally self-interested but systematically so. Where you give once and lose, give again and lose again, and the pattern is not a misunderstanding.
Axelrod’s later analysis looked at this. A population of pure defectors, he found, is actually stable — no single cooperator can survive in it. A lone Tit for Tat player in a room full of defectors will lose every round.
But here is the thing: they do not have to be alone.
When a small cluster of cooperators find each other — even in a hostile environment — they begin to outperform the defectors over time. The cooperators do well with each other. The defectors do well against isolated cooperators, but among themselves they do poorly. Over enough rounds, cooperation spreads.
The implication is practical. In a hostile environment, the first move is not to change your strategy. It is to find your cluster. The people who play the same way you do. The relationships where cooperation goes in both directions.
You cannot force an environment to be different by being nice to it. But you can protect yourself from it by not being isolated within it.
There is one more finding worth sitting with.
Axelrod’s research showed that cooperation emerged more naturally in environments where people expected to keep interacting — what he called the “shadow of the future.” The longer the game, the more rational cooperation becomes.
But the more useful reading of this is not about engineering circumstances. It is about your own consistency.
The four properties — nice, provocable, forgiving, clear — only work as a strategy if they are applied consistently. A player who is nice sometimes, and forgiving sometimes, and retaliates inconsistently, sends no legible signal. The other party cannot cooperate with a strategy they cannot read.
This means the real work is not situational. It is dispositional. You decide in advance how you will respond to cooperation and how you will respond to defection, and you hold to it regardless of mood, context, or who is watching. Not because it will always produce the outcome you want in any given interaction, but because it makes your behaviour something other people can rely on.
Being nice is a choice you make before you know how the other person will behave. Being forgiving is a choice you make after they have already cost you something. Ensuring that defection has a consequence is a choice you make even when it would be easier to let it pass.
None of these are automatic. They require a decision each time. The framework does not remove the judgement call — it just gives you a principled basis for making it.
I am not naturally strategic in the calculating sense.
I do not think about interactions as games. I do not model what the other person is likely to do and optimise accordingly. That framing makes most of life feel exhausting and thin.
But what I took from Axelrod’s work was not a strategy. It was a permission.
Permission to be open first, without that being naive. Permission to respond when something goes wrong, without that being aggressive. Permission to forgive quickly, without that being weak. Permission to keep things simple, without that being unsophisticated.
The mathematics is clear on this: being nice is not the same as being soft. Being forgiving is not the same as being a pushover. The strategies that win over time are not the ones that extract the most from any single encounter. They are the ones that other players want to keep playing with.
That is not a bad principle for most things.
Start open. Respond clearly when needed. Forgive quickly. Stay simple enough to be understood.
Then find the people who play the same way.
The rest tends to take care of itself.