Page 26 -

2 Historical Introduction 19

believe that (3) if neither confesses, both will go clear” (Poundstone 1992, pp. 117–
118). In the non-iterated version, the rational solution is that both confess—but if
they believe they can trust each other, they can both win, as both will go clear if
neither confesses. Axelrod’s question was under which conditions a prisoner in
this dilemma would “cooperate” (with his accomplice, not with the police) and
under which condition they would “defect” (i.e. confess, get a reward and let
the accomplice alone in prison). Super-strategies in this tournament had to define
which strategy—cooperate or defect—each player would choose, given the history
of choices of both players, but not knowing the current decision of the partner.
Then every strategy played the iterated game against every other strategy, with
identical payoff matrices—and the tit-for-tat strategy proved to be superior to 13
other strategies proposed by economists, game theorists, sociologists, psychologists
and mathematicians (and it was the strategy that had the shortest description in
terms of lines of code). Although later on several characteristics of several of
the strategies proposed could be analysed mathematically, the tournament had at
least the advantage of easy understandability of the outcomes—which, by the way
is another advantage of the “third symbol system” over the symbol system of
mathematics.
Cellular automata later on became the environment of even more complex
models of abstract social processes. They serve as a landscape where moving,
autonomous, proactive, goal-directed software agents harvest food and trade with
it. Sugarscape is such a landscape which serves as a laboratory for a “generative
social science” (Epstein and Axtell 1996, p. 19) in which the researcher “grows”
the emergent phenomena typical for real-world societies in a way that includes the
explanation of these phenomena. In this artificial world, software agents find several
types of food which they need for their metabolism, but in different proportions,
which gives them an incentive to barter with a kind of food of which they have
plenty, for another kind of food which they urgently need. This kind of a laboratory
gives an insight under which conditions skewed wealth distributions might occur or
be avoided; with some extensions (König et al. 2002), agents can even form teams
led be agents who are responsible to spread the information gained by their followers
among their group.

2.4 Conclusion and Suggested Further Reading

This short guided tour through early simulation models should have shown the
optimism of the early adopters of this method: “If it is possible to reproduce, through
computer simulation, much of the complexity of a whole society going through
processes of change, and to do so rapidly, then the opportunities to put social science
to work are vastly increased” (Ithiel de Pool and Abelson 1961, p. 183). Thirty-five
years later, Epstein and Axtell formulate nearly the same optimism when they list
a number of problems that social sciences have to face—suppressing real-world
agents’ heterogeneity, neglecting nonequilibrium dynamics and being preoccupied