Much of human interaction is shaped by the structure of the prisoner’s dilemma. We put in place institutions and norms to enforce cooperation. We tell shared stories to inspire it. We evolved moral emotions to achieve cooperation on an interpersonal level: empathy and gratitude to assure cooperators of our cooperation, anger and vengefulness to punish defectors, tribalism and loyalty to cooperate with those we know well.
But the crux of the prisoner’s dilemma is that defection is always better for the defector. We try to get others to cooperate with us, but we also try to defect as much as we can get away with. We want our peers to pay their taxes, admit mistakes, share credit, and stay faithful. We also fudge our taxes, shift blame, boast, and cheat.
There are many strategies for dealing with PD, and some of them can be formalized in code and entered in competitions with other strategies. The simple strategies are named and studied: tit-for-tat responds to each play in kind, tit-for-two-tats forgives one deception in case it was a mistake, Pavlov changes tacks after being defected against and so on and so forth. Which strategy works best?
It turns out that the success of each strategy depends almost entirely on the strategies played by opponents. Each approach can fail to reach cooperation with others or under-exploit opportunities to defect; even a strategy of always defecting is optimal if enough other players always cooperate. If you only knew what your opponent is playing, you could always choose the best option.
And this brings us back to predicting other humans. If we can model their strategies, if we know who will be forgiving and kind, who will be vengeful and dangerous, we can play optimally in any situation. Predicting well is the unbeatable strategy.