In 1984, political scientist Robert Axelrod ran a tournament: any researcher could submit a strategy for playing an iterated Prisoner’s Dilemma. Elaborate, clever, machine-learning-style strategies were entered. The winner was four lines of code called Tit-for-Tat: cooperate first, then do whatever the other player just did. It beat everything (Axelrod & Hamilton, 1981).
That tournament is probably the single best answer to the question how did humans ever trust each other.
The core idea
Biologist Robert Trivers asked in 1971: how can altruism evolve between non-relatives? Helping a stranger seems like a fitness-losing move.
His answer was reciprocal altruism: I help you now, you help me later, and if we both do this repeatedly, we both come out ahead (Trivers, 1971).
Three conditions make it work:
- Repeated interactions. One-shot encounters favor defection.
- Individual recognition. You have to know who helped and who didn’t.
- Punishment of cheaters. Defectors lose access to future cooperation.
When all three hold, cooperation between strangers becomes evolutionarily stable. When any of them breaks, cooperation collapses.
Tit-for-Tat, and why it wins
Axelrod ran his tournament again and again with different entrants. Tit-for-Tat kept winning. The reasons, from his analysis:
- Nice: never defects first
- Retaliatory: punishes defection immediately
- Forgiving: returns to cooperation the moment the other player does
- Clear: simple enough that other players can predict it and cooperate
The combination is what works. Pure cooperators get exploited. Pure defectors get isolated. Complex strategies confuse partners. Nice + firm + forgiving beats all of them.
A slight upgrade called Generous Tit-for-Tat occasionally forgives a defection it shouldn’t, which helps when partners are noisy or make mistakes. In messy real environments this beats strict Tit-for-Tat.
The most robust social strategy ever found is: cooperate first, mirror second, forgive third.
Your brain has a cheater-detector
Leda Cosmides and John Tooby ran a striking experiment with the Wason selection task. When you phrase a logical puzzle abstractly (“If a card has a vowel on one side, it has an even number on the other”), people get it wrong 75–90% of the time.
When you phrase the identical logic as a social contract (“If you drink beer, you must be over 18”), accuracy jumps to ~75% correct (Cosmides, 1989).
Same logic. Different framing. A giant jump in performance.
Cosmides’ interpretation: we didn’t evolve general-purpose logic. We evolved a specialized cheater-detection module that activates when the problem looks like “who’s violating a social exchange.”
This module is on whenever you’re evaluating a deal, a favor, a promise, a relationship.
We pay to punish cheaters
Ernst Fehr and Simon Gächter set up a public goods game where anonymous strangers played together once, knowing they’d never interact again. When they added the option to pay real money to punish defectors (who you’d never see again, no reputation benefit), people used it heavily.
Cooperation rates jumped from ~30% to ~80% once punishment was available (Fehr & Gächter, 2002).
People paid from their own pocket to punish strangers for defections that didn’t even happen against them. This is “altruistic punishment” and it’s one of the strongest forces keeping human cooperation stable at scale.
How this gets exploited
The reciprocity circuitry was calibrated for small groups where everyone knew each other. Modern life ignores most of those assumptions.
- One-shot encounters masquerading as repeated ones. Door-to-door scammers, cold callers, pickup artists. They trigger reciprocity circuitry knowing there’s no round two.
- Unsolicited favors create social debt. The free pen, the complimentary drink, the “I’ll just do this one thing for you.” Your brain logs it as an obligation whether you want it to or not (Cialdini, 2006).
- Fake reputation layers. Sock puppet reviews, astroturfed endorsements, borrowed credentials. Indirect reciprocity relies on reputation signals, and those are cheap to forge online.
- Manufactured generosity. “I’m doing you a favor” as a sales lead-in. The favor is real. The relationship isn’t.
Why this matters
- One-shot exploiters hack a system built for repeated games. If they’ll never see you again, reciprocity is the wrong algorithm.
- Failing to retaliate teaches people you’re safe to farm. Pure cooperators get exploited. Retaliation is part of the strategy, not a personality flaw.
- Keeping score too precisely kills relationships. Generous Tit-for-Tat beats strict Tit-for-Tat. Round up, not down.
- You have cheater-detection hardware; use it. If your brain says “something’s off about this deal,” the specialized module is firing. Trust it.
Cooperate first. Mirror second. Forgive third. Punish when needed. This is the oldest working recipe for trusting strangers.