6 Agent-Based Modeling and Simulation with Swarm
periment, under spe c ific conditions, human-like behavior can be fabricated by
both humans a nd machines if appropriate formal rules are provided. Searle
therefore argues that strong AI is impossible to realize.
Various counterarguments have been considered in response to Searle, and
questions that would probably occur to most people include
• Can conversion rules be w ritten for all possible inputs?
• Can such an immense database actua lly be searched?
However, these counterarguments ar e devoid of meaning. The former rejects
the realization of AI in the first place, and the latter cannot be refuted in
light of the possibility that ultra-high-speed parallel computing or quantum
computing may exis t in the future. Thus, neither one can serve as the basis
of an argument.
One powerful counterargument is based on system theory. Altho ugh the
person in the room certainly lacks understanding, he constitutes no more than
a single part of a larger sy stem incorporating other elements, such as the paper
and the database, a nd this system as a whole does posses s understanding. This
point is integral to the complex sys tems regarded in this book. The level at
which intelligence is sought depends on the observed phenom e non, and if the
phenomenon is c onsidered as being an emergent property, the validity of the
above system theory can be recognized. Moreover, a debate is ongoing about
whether intelligence should be thought of as an integrated c oncept or as a
phenomenon that is co-evolving as a result of evolution.
1.3 Criticism of simulation
It should be kept in mind that a simulation is not an omnipotent tool,
which is reflected in the pilots example (see the quote at the beginning of this
chapter). The limitations of simulation are the subject of lengthy discuss ions
in the field of robotics.
The ultimate goal in robotics is setting actual machines in motion. How-
ever, the process of enabling rob ots to move is costly, and its implementa-
tion is not straightforward. Thus, simulation is actively used for exp e rimental
purposes, and an increasing number of studies employ only simulation for
conducting experiments, without any verification using ac tual machines. In
conducting research on humanoid ro bots at our laboratory, an elaborate sim-
ulator is always prepared as a pre liminary experiment (see Figs. 1.4 and 1.5),
and movements realized in the simulator are often impo ssible to perform with
an a c tual robot. The primary reasons for performing these simulations in-
clude loc ating unforeseen sensor errors, monitoring fatigue due to prolonged
Introduction 7
FIGURE 1.4: Simulation of humanoid robot motions.
FIGURE 1.5: Real-world motions of a humanoid robot.
use, and studying differences in the friction with the floor. However, the q ues-
tion is whether it would ever be possible to simulate all these circumstances
a priori.
In 1991, Rodney Brooks, an eminent scholar in the AI field who propose d
concepts such as “intelligence without representation” and “subsumption ar-
chitecture,” presented a highly intriguing argument at the International Joint
Conference on Artificial Intelligence (IJCA91). He stipulated that to faithfully
recreate the real world in a simulation is impossible, and on the basis of this
he emphasized the importance of the following in AI research.
Physical grounding This hypothesis states that intelligence should be
grounded in the interaction be tween a physical agent and its environ-
ments.
Embodiment General intelligence cannot be produced in a vacuum. Com-
8 Agent-Based Modeling and Simulation with Swarm
puters with human intelligence should have a solid sensor-motor base
upon which higher cognitive functioning can be built (or evolved).
This statement also serves as a warning regar ding both simulation techniques
and AI arguments that are likely to become purely theoretical abstractions.
With this in mind, simulation experiments must be performed with extreme
care.
1.4 Swarm and the Santa Fe Institute
We now present a description of Swarm, which is a multi-agent simulation
library developed at the Santa Fe Institute, which is r e nowned for its research
on complex systems. Since Swarm suppor ts Java and Objective-C, it can be
easily utilized in object-oriented modeling of phenomena. More specifically,
Swarm allows for str aightforward implementation of the following functions,
each of w hich is ex tremely useful in complex systems research.
• Interactive acc e ss to fields and methods of objects in the simulation
• Graphical representation of different aspects of the simulatio n (such as
the distribution of age nts on a two -dimensional surface and various sta-
tistical markers, which can be represented through linear graphs and
histograms)
• Assignment of an independent clock (scheduler) to each layer of the
simulation
The Santa Fe Institute, which acts as the headquarters for complex systems
research, is introduced below. Santa Fe is the capital o f the U.S. state of New
Mexico, and is located at the center of the state. New Mexico contains a num-
ber of settlements known as pueblos that are, even now, populated by Native
Americans. Santa Fe is a city that is becoming increasingly popular as a travel
destination for American tourists, mainly owing to the preserved landscapes
and buildings reminiscent of old times. The famo us Mesa Verde National Park
is also located nearby, where the esca rpments reveal a large number of ruins
left behind by a mysterious people, the Anasazi Indians (Fig. 1.6). This tribe
thrived as a highly developed civilization about 1400 years ago, and then 700
years later vanished suddenly. This process of thriving and demise is a re search
topic in Swarm simulations.
The Santa Fe Institute was originally constructed in the ruins of an abbe y
as part of the Los Alamos Nationa l Laboratory. A number of pr ominent schol-
ars worked at this institute, including Kenneth Arrow (winner of the No-
bel Prize in Economics), Murray Gell-Mann (winner of the Nobe l Prize in
Introduction 9
FIGURE 1.6: Mesa Verde National Park.
Physics for his work on quarks), Robe rt Axelrod (prisoner’s dilemma; see Sec-
tion 4.3.2), John Holland (the creator of genetic algorithms), and Chr istopher
Langton (who proposed the concept of artificial life).
The “general impossibility theorem” proposed by Arrow is outlined here,
since this theorem is related to the concept of complex systems. In this theo-
rem, three bored students (A, B, and C) are discussing which of the following
three e ntertainment options they should choose.
• Movie
• Television
• Karaoke
Let us assume that the order of preference for each per son is as follows:
Preference order for A Movie>television>karaoke
.
Preference order for B Televis ion>karaoke>movie
.
Preference order for C Karaoke>movie>television
.
They decide to determine their choice democratically and ad opt a majority
vote. First, in deciding between a movie and TV, the following distribution
determines tha t a movie is preferable.
A movie is preferable to television: A and C.
Television is preferable to a movie: B.
10 Agent-Based Modeling and Simulation with Swarm
∴ Movie (two votes) > television (one vote).
Next, in deciding between television and karaoke, the distribution yields tele-
vision as the winning choice:
Television is preferable to karaoke: A and B.
Karaoke is preferable to television: C.
∴ Television (two votes) > karaoke (one vote).
From the above results, the order of preference is as follows:
Movie > television > karaoke.
Accordingly, the final decision is to watch a movie. Howe ver, looking at the
preferences for a movie and karaoke, the distribution leaves karaoke as the
winning choice:
A movie is preferable to karaoke: A.
Karaoke is preferable to a movie: B and C.
∴ Karaoke (two votes) > movie (one vote).
In other words, the two results contradict each other.
Arrow generalized this result and demonstrated that serious contradictions
arise even in cases with a larger number of people. For example, by considering
the relation between the preferences of 100 peo ple, the following result can be
obtained by skillfully manipulating the order of preference in a majority vote.
• From a number of majority votes, the overall conclusion is that “x is
preferable to y.”
• However, only one person prefers x over y.
• The remaining 99 people prefer y over x.
Arrow’s general impossibility theorem proved that a “democratic decision”
that satisfies the following four criteria does not exist.
1. All preference orders are allowed.
2. Citizen s overeignty is assumed. In other words, x is selected when e v-
eryone agrees that x > y.
3. The preference order b e tween two choices depends only on the individual
preferences, and is not affected by other alternatives.
4. There is no dictator.
The third criterion is called “independence of irrelevant alternatives,” and
is an important assumption. The following is a situation where this assumption
does not hold.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.