Prisoner's Dilemma

In rationalist circles, you’ll often see reference to a thought experiment called the “prisoner’s dilemma.” It posits a situation wherein police are interrogating two members of criminal gang about an offense that carries a sentence of ten years in prison. The police have a problem in that they don’t have the evidence to convict on the higher charge if neither of the suspects confesses. But they can sentence them both to two years on a lesser charge, and they use this to offer each a bargain. If ones confesses and rats out the other, he’ll get to go free while the other gets the entire ten-year prison term. If they both stay quiet, they’ll each get two years, and if they rat out each other at the same time, they split the max sentence and get five years each.

So the scenario has four possible outcomes, and for the sake of clarity let’s label the two criminals Suspect A and Suspect B:

• Both of the suspects keep their mouths shut, and so both get a two-year prison stretch

• Suspect A confesses before Suspect B, and so gets to go home a free man while Suspect B gets ten years in prison

• Suspect B confesses before Suspect A, and so gets to go home a free man while Suspect A gets ten years in prison

• Both rat each other out and so both get five years in prison

Without getting into the specific game theory dynamics at play (much of which I can’t claim to fully understand), the gist is that the only rational result is that the prisoners betray each other when offered substantial incentive. The individual players in the game will act according to their self-interest, and will respond to carrots and sticks in line with this fact. The dynamics at play have been applied to several fields of human interaction, perhaps most successfully in economics, where it’s been affixed alongside interest behavior theories like that of schizophrenic mathematician John Nash.

The prisoner’s dilemma is a neat and tidy little experiment, cooked up by eggheads at the RAND Institute during the Cold War. And within its neatness lies the problem of “Blank Slatism.” Aside from its idealistic limitations — it assumes rational actors across the board. It takes place within a hermitically sealed scenario environment, with blank slate players stripped of all unique characteristics that could influence their behavior aside from pure rational self-interest. For instance, loyalty to one’s partner in crime is considered irrational, as are most religious considerations that could lead one to make decisions seemingly injurious to themselves in this life. And what do the prisoners know about each other? What is their history with each other?

It also ignores outside manipulation. The mafia essentially smashed the game by force through their strict Omertà code of silence, and made it known that if you did snitch and get your freedom, your free life would be a short one. They essentially froze the game at an optimum state — each prisoner keeps their mouth shut and they both get the two-year sentence. The three other options are wiped off the board given the outside threat, and so the mafia added their own incentive structure to that of the police. Human systems are always vulnerable to external attack, even the rational ones.

People with higher IQs than mine have developed and tweaked thought games like the prisoner’s dilemma, and they’ve striven to adapt them to the real world in creative ways that allow for more lawless variables and outside pressures. But I just find myself suspicious of “rationalist” dogmas about human behavior, especially when they’re used by midwits to predict the future in those most consequential of games, like politics and war.

This entry may seem like a rejoinder against my last post of vengeance. But I believe that while revenge can be a game theory-friendly balancing force, it’s also more primordial than simple rational “self-interest.” More adaptable to the big games. But of course it also has its limits in predictive models. The Cold War could have gone hot at any number of inflection points. There’s always that most mercurial of players — luck.