The Road to Agent-based Models
Reality is
complex, but models don't have to be. Sugarscape, for example, is a "bottom-up,"
agent-based model. The model itself is not complex; complex interactions
result from the aggregate result of simple agents' myriad interactions.
The concept
itself is simple enough. Actually reaching this conceptual point took
a while. Complex mathematical models have been, and are, common; deceivingly
simple models only have their roots in the late forties, and took the
advent of the microcomputer to really get up to speed.
1) Von Neumann
machines
Although
any discussion of the history of agent-based modeling must begin with
von Neumann machines, von Neumann machines are not agent-based modelsor, in fact, models of any kind at all. They are theoretical constructs,
but they represent a paradigm shift that laid the ground for the study
and construction of artificial life, and later bottom-up modeling.
In
the late 1940s, mathematician John
von Neumann suggested a machine capable of reproduction. The device
he proposed would follow precisely detailed instructions to fashion a
copy of itself. It would then gift the copy with a copy of the instructions,
and thus create a fertile offspring. Actually building a von Neumann machine
would have been a staggering task. But von Neumann's friend Stanislaw
Ulam, also a mathematician, suggested that the machine be built on paper,
as a collection of cells on a grid. The idea intrigued von Neumann, who
drew it upcreating the first of the devices later termed cellular
automata.
Although
von Neumann never finished his mathematical proof, the implications were
clear: the basis of life was information.
The concept
that life can exist as information led to some fascinating avenues of
research. One was John Horton Conway's game "Life." Unlike von Neumann's
machine, Life operated by tremendously simple rules:
Life occurs
on a virtual checkerboard. The squares are called cells. They are in
one of two states: alive or dead. Each cell has eight possible neighbors,
the cells which touch its sides or its corners.
If a cell on the checkerboard is alive, it will survive in the next
time step (or generation) if there are either two or three neighbors
also alive. It will die of overcrowding if there are more than three
live neighbors, and it will die of exposure if there are fewer than
two.
If a cell on the checkerboard is dead, it will remain dead in the next
generation unless exactly three of its eight neighbors are alive. In
that case, the cell will be "born" in the next generation.
Conway and
his team ran game after game, making every alteration of every cell by
hand, studying a variety of configurations. Later, Conway opened up the
game to competition, offering a prize for anyone who could prove that
Life was capable of generating an infinite population given a finite initial
configuration. The winning team programmed the game into a computerthe first known use of electronic computers to create artificial life.
Although
current research deals with larger questions, Life still has devotees,
with a number of web pages devoted to it. The simulation is a frequent
topic of discussion. Its effect on the path to agent-based modeling is
profound.
The life
on Conway's gameboard was pure information, the rules incredibly simplebut the implications were complex. And, it would lead others to wonder,
if such complex movements can result from such simple rules, might not
real-world behaviors be the result of surprisingly simple behavior?
The use
of artificial life to create bottom-up models of the real world followed
from there. Sugarscape is such an enterprise.
Another, earlier one was Craig Reynolds's "boids," short for "birdoids."
To the human observer, bird flocking appears to follow some centralized
control due to the sheer size and complexity of the flock.
Reynolds
suspected that flocking was a decentralized activity, and created a computer
model to investigate his theory. His agents, which he called boids, each
followed a few simple directives; the behavior that emerged was eerily
like flocking in nature. So much so, in fact, that the boids were used
as a starting point for computer animators who needed to animate flocks
for Hollywood films.
For more
about boids, see Craig Reynolds's boids
page.
Or go
here
to find links to more information on agent-based modeling.
Center on Social and Economic Dynamics The Brookings Institution, 1775 Massachusetts Ave NW, Washington DC 20036 (202) 797-6403 · Fax (202) 797-6144 · E-mail