Outwitting matched opponents: Natural selection
for agent-based modeling
The problem of anticipating the behavior of others is
evolutionarily ancient, and forms the core selective pressure for cognitive
adaptations in all animals.
Such adaptations typically take into account the recursive
nature of the causal networks that characterize social interaction; in their
earliest forms, they are cue-based response systems rather than simulations.
With the advent of mammals, cognitive adaptations undergo
a set of dramatic innovations, including play and dreaming. Play is an adaptation
for behavioral simulations, while dreaming involves mental simulation. Both
have features that suggest they were designed to train skills for high-stakes
adversarial encounters.
In this talk, I present research into playful interactions
between parents and children to demonstrate the characteristics of common
mammalian adaptations for agent modeling. I argue that human beings have
additional adaptations for waking mental simulations, and that human behavior
is governed to a significant degree by such simulations. A central feature
of human social interaction, therefore, is recursive mental simulations that
attempt to model the simulations of others.
Nature's machinery for running agent-based modeling is
highly sophisticated, but the knowledge that is produced is typically local
and quickly dated. I argue for the virtues of this approach.
Francis F. Steen
University of California in Los Angeles
Communication Studies
http://cogweb.ucla.edu
steen@commstds.ucla.edu