Cooperation evolves in computer tourney.

Evolutionary theorists often point out that natural selection -- the perpetuation of genetically based traits that increase an individual's chances of surviving and producing offspring -- favors selfish behavior. However, computer tournaments that pit a variety of strategies for obtaining payoffs in social encounters against one another provide clues to how natural selection also promotes cooperation among genetically unrelated individuals, scientists report in the Jan. 16 NATURE.

This brand of cooperation, known as reciprocal altruism, prevails when some individuals forgo constant selfishness for a "tit-for-tat" tactic, in which they cooperate with a colleague on a first encounter, and on subsequent occasions do whatever their cohort did on the previous encounter. In computer simulations, programs that eschew cooperation nearly wipe out their tit-for-tat compatriots, which then stage a comeback and give way to a "generous" version of tit-for-tat that forgives a new selfish deceptions by others, report zoologist Martin A. Nowak of the University of Oxford, England, and mathematician Karl Sigmund of the University of Wien, Austria.

Nowak and Sigmund tested tactics derived from the "prisoner's dilemma," in which two players score varying amounts of points by either cooperating or acting selfishly. An individual receives the most points by acting selfishly when the other cooperates; if both cooperate, each obtains a moderate payoff; and both score poorly in cases of dual selfishness. Prior computer tourneys suggested that, in repeated encounters, the tit-for-tat strategy outscores all others.

To better simulate encounters between biological organisms. Nowak and Sigmund programmed occasional random errors into each of 100 different prisoner's dilemma strategies. When the sample excluded the tit-for-tat approach, a strategy of consistent selfishnes gained dominion, because competing strategies did not immediately retaliate against exploiters. Although tit-for-tat initially performs poorly against selfish programs, "the tide turns when 'suckers' are so decimated that exploiters can no longer feed on them," the researchers assert. Reciprocators then eliminate exploiters and "generous" tit-fot-tat takes over.

The computer results suggest that evolution favors simple, probabilistic rules of conduct, writes biologist H.C.J. Godfray of Imperial College in Berkshire, England, in an accompanying editorial. Cooperative strategies may succeed more easily in nature than in computers, Godfray adds, since most animals interact with relatives more often than with strangers.