Page 25 -

18 K.G. Troitzsch

the “third symbol system” in social science proper was not introduced before the
first multilevel models and cellular automata that integrated at least primitive agents
in the sense of software modules with some autonomy.
Cellular automata (Farmer et al. 1984; Ilachinski 2001) are a composition
of finite automata which follow the same rule, are ordered in a (mostly) two-
dimensional grid and interact with (receive input from) their neighbours. The
behavioural rules of the individual cells are quite simple in most cases; they have
only a small number of states among which they switch according to relatively
simple rules, as in the famous game of life (Gardener 1970), where the cells have
only two states, alive and dead, and change their states according to the two simple
rules: if the cell is alive, it remains in this state if it has exactly two or three live cells
among its eight neighbours—otherwise it dies—and if the cell is dead, it bursts into
life if among its eight neighbours there are exactly three live cells. The great variety
of outcomes on the level of the cellular automaton as a whole enthused researchers
in complexity science and lay the headstone for innumerable cellular automata in
one or two dimensions.
One of the first applications of cellular automata to problems of social science
is Thomas Schelling’s (1971) segregation model, demo versions of which are
nowadays part of any distribution of simulation tools used for programming
cellular automata and agent-based models—a model that shows impressively that
segregation and the formation of ghettos is inevitable even if individuals tolerate a
majority of neighbours different from themselves.
Another example is Bibb Latané’s dynamic social impact theory with the imple-
mentation of the SITSIM model (Nowak and Latané 1994). This model, similar
to Schelling’s, also ends up in clustering processes and in the emergence of local
structures in an initially randomly distributed population, but unlike Schelling’s
segregation model (where agents move around the grid of a cellular automaton
until they find themselves in an agreeable neighbourhood), the clustering in SITSIM
comes from the fact that immobile agents change their attitudes according to the
attitudes they find in their neighbourhood and according to the persuasive strength
of their neighbours.
Other cellular automata models dealt with n-person cooperation games and
integrated game theory into complex models of interaction between agents and their
neighbourhoods, and these models, too, usually end up in emergent local structures
(Hegselmann 1996).
And in another computer simulation related to game theory run by Axelrod, it
could be shown that the tit-for-tat strategy in the iterated prisoner’s dilemma was
superior to all other strategies which were represented in a computer tournament
(Axelrod 1984). The prisoner’s dilemma had served game theorists, economists
and social scientists as a prominent model of decision processes under restricted
knowledge. The idea stems from the early 1950s, first written down by Albert
Tucker, and is about “two men, charged with a joint violation of law, are held
separately by the police. Each is told that (1) if one confesses and the other does
not, the former will be given a reward ::: and the latter will be fined ::: (2) if
both confess, each will be fined ::: At the same time, each has good reason to