WOULD BE WORLDS;
HOW SIMULATION IS
CHANGING THE
FRONTIERS OF SCIENCE
by John L. Casti
New York: John Wiley & Sons, Inc.
1997
Reviewed by: Darryl A. Baker
November 1996
In Would-Be Worlds, John Casti probes the world of simulation:
recreating complex real world situations using advanced computer applications.
The traditional method to study any science is the scientific method: using
controlled, laboratory experiments to form and test hypotheses. However,
there are circumstances where it has been difficult if not impossible to
determine effects of a certain action upon a particular environment. Issues
such as global warming, monetary policy, stock market fail safe mechanisms,
genetic engineering, and medical vaccination research require predictive
models to determine success. But in such critical day to day functions,
such as the stock market, which can have a huge impact on the global economy,
controlled experiments are next to impossible. Real life complex systems
involve agents that are "drivers, traders, molecules and are both
intelligent and adaptive" (p. x). They act upon a set of rules but
modify the rules with the input of new information. In sporting events
like football, or financial market trading, the agents (players, traders)
interact with each other, learn from the action taking place and modify
their strategy based upon those events. Continuing development of simulation
tools using computer technology can significantly improve the predictive
process in these environments: "With our newfound ability to create
worlds for all occasions inside the computer, we can play myriad sorts
of what-if games with genuine complex systems" (p. xi). It is Mr.
Casti's intent to examine various complex computer systems that attempt
to solvereal world problems using simulation or artificial worlds.
Casti first examines the notion of artificial life or biomorphs.
This
is a program that was first developed at the Los Alamos National Laboratory
in cooperation with the Sante Fe Institute and Apple Computer Corporation.
Pioneering work by biologist Richard Dawkins, Cliiford Pickover, and Aristid
Lindenmayer helped create systems that produced very real looking yet primitive
two-dimensional organism images with a computer, using a set of procedures
and mathematical rules. The images produced are mathematical objects that
resemble forms of living organisms. Casti poses the question of whether
these creations are "curiosities" or have actual biological relevance.
As in all of the computer models examined by Casti, he show restraint by
not declaring that the science is a monstrous breakthrough that will change
the world; He avoids a position that many early artificial intelligence
researchers fell into--making broad, overstated predictions of its use.
The key in evaluating the success of a simulation is to determine how well
the model answered the original question posed. The computer used a set
of rules that generated images, but that does not mean the same rules were
followed by nature in creating real flowers. Models can be built to "predict
or explain something about the real world process they represent."(p.
46) Computers can build worlds by generating patterns and ideas in bits
and bytes of ASCII (binary) codes.
Artificial Intelligence has roots in many disciplines including linguistics.
Casti refers to the linguistic research as it relates to computer programs
done by Norm Chomsky. Using sentence structure, music language and structure,
and art patterns, Casti suggests that these are also rule-based systems.
Computers can create worlds that mirror human creativity by using rule
based systems. A computer simulation can be drawn in two ways: top down
and bottom up. The difference is in the details. Bottom up simulations
try to give detailed technical information while top down simulations try
to make an "artistic impression." A top down simulation is strong
on processes but weaker on the realism of the data. It focuses on a broader
aspect of a real world situation and expoits that aspect in its simulations.
A real world simulation requires a high level of accuracy (detail), and
include social inhibitions and psychological constraints. The bottom up
simulations that Casti refers to are more representative of real world
situations. An example of a top down simulation is a commercial "war"
game.
Bottom up examples include models for testing stock market behaviors. Casti
introduces several working applications of real world simulations in Would-Be
Worlds. One such world that may have an actual use in solving traffic
and air pollution problems is TRANSIMS. This system contains geographic
information of a city, roads, households, types of vehicles used, locations
where people shop, work, study, play, when they want to go there, traffic
light patterns and layouts of one way streets. It then uses a procecure
to compute a "good route" to get the commuter to the destination
using preferred roads and minimizing travel time. The system then starts
the process of actually moving people through the traffic. In each additional
step, new calculations are entered accounting for congestion and accidents.
Vehicles are coded for "operating characteristics" such as engine
type, exhaust systems, tire pressure and speed to measure environmental
impact. The module attempts to answer the question "How does a given
proposed change in the system (road contruction, high-speed rail construction)
create traffic patterns and how do these patterns impact the environment."
(p. 134). "What-if" scenarios are introduced to the system to
simulate the reaction of traffic patterns and other variables. Casti suggests
that these types of models can have a dramatic impact on decision making
processes.
Other systems Casti investigates are more sophisticated and lean towards
actual learning methods. Corewar is a game in which two or more
computer programs "stalk" each other and set out to destroy one
another. Using sophisticated programming, the programs "learn"
where the other programs memory is located in order to destoy it. These
systems attempt to emulate the same activity and structure that are present
in the human brain. This is done with the use of neural networks and genetic
algorithms. A system called Tierra was developed that created an
"ancestral organism capable of replication and open-ended evolution."
(p. 163). Tierra uses the same structure of so called digital life of a
computer that goes through a process whereby programs are allocated time
and memory to operate. The programmer "created a set of software instructions
that mimic the operation of a physical hardware machine." (p. 164).
The organisms were able to evolve into "social" creatures into
groups of closely related organisms.
Unlike other, earlier research in artificial intelligence, John Casti
realizes the limitation of the field of simulation. Many predictions of
artificial intelligence machines were vastly exaggerated. Casti reports
on very successes and limitation of several types of simulation machines.
The excitement of breakthrough technologies and their potential impact
upon humans is also tempered by the realization that we still have much
more to learn and to research in simulation techniques. Anyone interested
just a little in virtual reality or computer games would find this book
fascinating and informative without being overly technical.
Related Web Sites:
The Sante Fe Institute
Los Alamos National
Laboratory
Center for the
Neural Basis of Cognition
TRANSIMS
Tierra
Sugarscape
Swarm
Simulation System
Image Credits:
Hawkseed flower using L-system rules. P. Prusinkiewics, Mark Hammel,
Eric Mjoiness.
© 1993 Association for Computing Machinery
TRANSIMS Photo, Los Alamos National Laboratory (p.140 in Would-Be
Worlds)
Created
by Darryl A Baker
November 8, 1996
Please send comments to dbaker@gslis.utexas.edu
Go back to: Titles of Book Reviews | AI and Expert Systems | Tentative Calendar | GSLIS
Page maintained by H. Haubitz
Last updated: November 17, 1996