The Prisoner's Dilemma by Carl Sagan

Prisoner's Dilemma
"A New Way To Think About Rules To Live By"

by Carl Sagan, Parade magazine, 28 Nov 1993

Moral codes that seek to regulate human behavior have been with us not only since the dawn of civilization but also among our pre-civilized, and highly social, hunter-gatherer ancestors. And even earlier. Different societies have different codes. Many cultures say one thing and do another. In a few fortunate societies, an inspired lawgiver lays down a set of rules to live by. But many revered codes have failed to establish a long-lived moral order. For example, the codes of Ashoka (India), Hammurabi (Babylon), Lycurgus (Sparta) and Solon (Athens), which once held sway over mighty civilizations, are today largely defunct. Perhaps they misjudged human nature and asked too much of us. Perhaps experience from one epoch or culture is not wholly applicable to another.

In this article, I describe an early effort - tentative but emerging - to approach the matter scientifically.

In our every day lives, as in the momentous affairs of nations, we must decide: What does it mean to do the right thing? How do we deal with an enemy? Should we ever take advantage of someone who treats us kindly? If hurt by a friend, or helped by an enemy, should we reciprocate in kind?

Examples are all around us: Your sister-in-law ignores your snub and invites you over for Christmas dinner. Should you accept? A co-worker makes you look bad in front of the boss. Should you try to get even? Should you cheat at cards? On a larger scale: Should we kill killers? If a power company supports a symphony orchestra, ought we to ignore its destructive, although legal, pollution of the environment? Shattering a worldwide voluntary moratorium, China resumes its testing of nuclear weapons. Should we?

In making such decisions, we're concerned not only with doing right but also with what works - what makes us and the rest of society happier and more secure. There's a tension between what we call ethical and what we call pragmatic. If, even in the long run, ethical behavior were self-defeating, we would not call it ethical, but foolish. (We might even claim to respect it but in practice ignore it.) Bearing in mind the variety and complexity of human behavior, are there any simple rules - whether we call them ethical or pragmatic - that actually work? Let's look at some of the rules we're taught:

THE GOLDEN RULE. The most admired standard of behavior in the West is the Golden Rule. Its formulation in the first-century Gospel of St. Matthew is: "Do unto others as you would have them do unto you." Almost no one follows it consistently. When the Chinese philosopher K'ung-Tzu (known as Confucius in the West) was asked in the sixth century B.C. his opinion of the Golden Rule - of repaying evil with kindness - he replied, "Then with what will you repay kindness?"

THE SILVER RULE. The Silver Rule is different: "Do not do unto others what you would not have them do unto you." The most inspiring 20th-century exemplars of the Silver Rule are Mohandas Gandhi and Dr. Martin Luther King Jr. They counseled oppressed peoples not to repay violence with violence, but not to be compliant and obedient either. Non-violent civil disobedience was what they advocated - putting your body on the line and showing, by your willingness to be punished in defying an unjust law, the justice of your cause. They aimed at melting the hearts of their oppressors. It worked, up to a point. But even Gandhi had trouble reconciling the rule of nonviolence with the necessities of defense against those with less lofty rules of conduct.

THE BRAZEN RULE. "Repay kindness with kindness," said Confucius, describing relations between individuals, "but evil with justice." This might be called the Bronze or Brazen Rule: "Do unto others as they do unto you." It's "an eye for an eye, and a tooth for a tooth," plus "one good turn deserves another." In actual human (and chimpanzee) behavior, it's a familiar standard. Without having to appeal to anyone's better nature, we institute a kind of operant conditioning, rewarding others when they're nice to us and punishing them when they're not. We're not pushovers, be we're not unforgiving either.

THE IRON RULE... AND OTHERS. Of baser coinage is the Iron Rule: "Do unto others as you like, before they do it unto you." It's sometimes formulated as, "He who has the gold makes the rules," underscoring not just its rejection of, but also its contempt for, the Golden Rule. This is the secret maxim of many, if they can get away with it, and often the unspoken precept of the powerful.

Finally, I should mention two mixed rules, found throughout the living world. They explain a great deal. One is: "Suck up to those above you, and intimidate those below." This is the motto of bullies. It's really the Golden Rule for superiors, the Iron Rule for inferiors. Since there is no known alloy of gold and iron, we'll call it the Tin Rule for its flexibility. The other common rule is: "Give precedence in all things to close relatives, and do as you like to others" - the Golden Rule for relatives, the Iron rule for others. This Nepotism Rule is known to evolutionary biologists as "kin selection."

Despite its apparent practicality, there's a fatal flaw in the Brazen Rule: unending vendetta. Each act of justifiable retribution triggers another. Violence begets violence. The reasonable part of us tries to keep the peace, but the passionate part of us cries out for vengeance. Extremists in the two warring factions can count on one another. They are allied against the rest of us, contemptuous of appeals to understanding an loving kindness. A few hotheads can force-march a legion of more prudent and rational people to brutality and war.

WHAT GAMES TEACH US. Clearly, the Brazen Rule is too unforgiving. But the Golden and Silver Rules seem too complacent. They systematically reward cruelty and exploitation. It is hard to imagine a Hitler or a Stalin being shamed into redemption by good example. The Iron Rule promotes the advantage of a ruthless and powerful few against the interest of the many. So is there a rule between the Golden and the Silver, on the one hand, and the Brazen and Iron, on the other, which works better than any of them?

Suppose we seek not to confirm or deny what we've been taught but to find out what really works. Is there a way to test alternative codes of ethics?

We're used to playing games in which somebody wins and somebody loses. Every point made by our opponent puts us that much farther behind. "Win-lose" games seem so natural that many people are hard-pressed to think of a game that isn't win-lose. In win-lose games, the losses just balance the wins - that's why they're also called "zero-sum" games.

Many children are appalled the first time they really come face to face with the "lose" side of win-lose games. On the verge of bankruptcy in the game Monopoly (tm), for example, they plead for special dispensation. When this is not forthcoming, they may, in tears, denounce the game as heartless and unfeeling - which, of course, it is. Within the rules of Monopoly, there's no way for players to cooperate so that all benefit. That's not how the games is designed. The same is true for boxing, football, hockey, basketball, baseball, lacrosse, tennis, racquetball, pinochle, chess, all Olympic events, yacht and car racing, potsy and partisan politics. There may be rewards for teamwork, but not for teamwork with the opponent. In none of these games is there an opportunity to practice the Golden or Silver Rule, or even the Brazen. There is room only for the Rule of Iron.

Nuclear war, however (and many conventional wars), economic depression and assaults on the global environment are all "lose-lose" propositions. Such vital human concerns as love, friendship, parenthood and the pursuit of knowledge are "win-win" propositions. Everyone gains from the creation of great music, art, architecture and literature, wise and just laws and, indeed, far-seeing moral codes. Our vision is dangerously narrow if all we know is "win-lose."

THE PRISONER'S DILEMMA. The scientific field that deals with such matters is called "game theory." It's used in military strategy, trade policy, corporate competition and the limiting of environmental pollution. The Defense Department has its own gaming agency. The paradigmatic game is the Prisoner's Dilemma. It is not zero-sum. Win-win, win-lose and lose-lose outcomes all are possible. It is wholly pragmatic and amoral:

Imagine that you and a friend are arrested for committing a serious crime. Before the two of you have any chance to compare stories or plan strategy, you are taken to separate interrogation cells. There, oblivious of your Miranda rights ("You have the right to remain silent..."), the police try to make you confess. They tell you, as police sometimes do, that your friend has confessed. The police might be telling the truth. Or they might be lying. If you're willing to say anything, what's your best tack to minimize punishment?

You're permitted only to plead guilty or not guilty; you cannot implicate or clear your friend. These are the possible outcomes:

* If you deny committing the crime, and (unknown to you) your friend also denies it, the case might be hard to prove. In the ensuing plea bargain, both your sentences will be very light.

* If you confess, and your friend does likewise, then the effort the State must expend to solve the crime is small. In exchange, you both will be given a fairly light sentence, although not so light as if you both had asserted your innocence.

* If you plead not guilty, and your friend confesses, the State will ask for a maximum sentence for you and minimal punishment (maybe none) for your friend. Uh-oh. You're very vulnerable to a kind of double cross. So's he.

So if you and your friend both plead innocent, you both escape the worst. But each must be sure of the other.

Should you play it safe and guarantee no worse than a middle range of punishment by confessing? Then, if your friend pleads innocent while you plead guilty - well, too bad for him, and you might get off scot-free.

When you think it through, you realize that, whatever your friend does, you're better off confessing. Maddeningly, the same holds true for your friend. But if both of you confess, you both are worse off than if both of you had pleaded innocent. This is the Prisoner's Dilemma.

Robert Axelrod, a professor of political science at the University of Michigan, has pioneered the study of a repeated Prisoner's Dilemma in which the two players go through a sequence of such games with no direct communication between them. At the end of each, they figure out from their punishment how the other must have pleaded. They gain experience about each other's strategy (and character). Will they learn to "cooperate" game after game - both always denying that they committed any crime - even if the reward for finking on the other (or "defecting") is very large?

If you cooperate overmuch, the other player may exploit your good nature. If you defect overmuch, your friend is likely to retaliate often, which will be bad for both of you. What is the right mix of cooperation and defection? How to behave then becomes, like any other question in Nature, a subject to be investigated experimentally.

This matter has been explored by Axelrod in a continuing round-robin computer tournament. Various codes of behavior confront one another, and at the end we see who wins (who gets the lightest cumulative prison term). The simplest strategies might be to cooperate all the time, no matter how much advantage is taken of you; or never to cooperate, no matter what benefits might accrue from cooperation. Both the Golden Rule and the Iron Rule always lose - the one from an excess of kindness, the other from an overabundance of ruthlessness. Strategies that are slow to punish defection lose, in part because they send a signal that non-cooperation works.

A RULE THAT WORKS. The most effective strategy in many such tournaments is called "Tit-for-Tat." It's very simple: You start out cooperating and, in each subsequent round, simply do what your opponent did last time. You punish defections, but once the other player cooperates, you're willing to let bygones be bygones. At first it seems to garner only mediocre success. But as time goes on, the other strategies defeat themselves - from too much kindness or too much cruelty - and this middle way pulls ahead.

Except for always being nice on the first move, Tit-for-tat is identical to the Brazen Rule. It promptly (in the very next game) rewards cooperation and punishes defection, and it has the great virtue that it makes its strategy absolutely clear.

To succeed, Tit-for-tat strategists must find others who are willing to reciprocate - players with whom to cooperate. Once there get to be several players employing Tit-for-tat, they rise in the standings together. After the first tournament, in which the Brazen Rule unexpectedly won, some experts thought it would pay to be less forgiving. Next tournament, they tried to exploit the Brazen Rule by defecting more often. They did poorly. Even experienced strategists tended to underestimate the power of forgiveness and reconciliation.

The superiority of the Brazen Rule in such tournaments was discovered by Axelrod and described in his remarkable book "The Evolution of Cooperation." A variant of Tit-for-tat that forgives other players for defecting occasionally - say 10 percent of the time - does even better if there's any chance of misunderstanding. We might call it the Goldplated Brazen Rule. Among other virtues, it breaks out of unending vendetta.

The Prisoner's Dilemma is a very simple game. Real life is considerably more complex. But its central lessons are striking: Be friendly at first meetings. Do not envy. Be generous; forgive your enemy if he forgives you. Be neither a tyrant or a patsy. Retaliate proportionately to an intentional injury (within the constraints of the rule of law). And make your behavior fairly (although not perfectly) clear and consistent. What would the world be like if more of us, individuals as well as nations, lived by these rules?