In the 1990s primate researchers became enthusiastic about the idea of Machiavellian intelligence—the ability that apes and humans have evolved for predicting and manipulating the behavior of other individuals. Apes and humans live in social groups where one's survival and reproduction prospects depend on one's social relationships. Once primatologists understood evolution from the selfish-gene viewpoint, they saw social interaction in a new light. Before, social behavior was thought to be for "pair-bonding" and "group cohesion." Now, it became viewed as a strategic game of politics, alliances, reciprocity, kinship, aggression, and peacemaking. A key to success in these strategic games is the ability to predict the behavior of other individuals. The Machiavellian intelligence theory suggests that great apes evolved larger brains and higher intelligence to better predict one another's behavior.
Suppose this view is right. Would evolution stop there, with everyone able to predict and manipulate everyone else's behavior? Or would counter-strategies evolve? In a society of Machiavellian psychoanalysts, individuals that are harder to predict and manipulate would have the usual protean advantages.
In their important 1984 paper on "mind-reading and manipulation," John Krebs and Richard Dawkins identified only two defenses an animal might use against having its actions predicted by an opponent: concealment and deception. You can try to hide your intentions (the poker-face strategy), or you can create a false impression about your intentions (the bluffing strategy). However, they overlooked the classic third option: randomness. The protean strategy. Doubtless each of these strategies is useful under particular conditions, and in a species with high Machiavellian intelligence, all of them would evolve. However, the protean strategy has one big advantage: it stops prediction dead in its tracks. The poker-face and bluffing strategies remain vulnerable to the evolution of better intention-sensing and deception-foiling abilities. But there is no way to improve prediction when you meet genuine randomness.
Was this article helpful?