Chapter 8
Conclusion
What is “real”? How do you define “real”? (from The Matrix
movie quotes)
This book discussed the simulation of complex systems from the basics
to implementation, applicable to a wide range o f fields. In particular, imple-
mentation in Swarm, a simulation suite for complex systems, was explained
in order to assist the reader to easily construct simulations. Complex systems
and artificial life are actively being researched, which finds a place in many
practical applications in the fields of animation and design, among others.
Therefore, the objective of this book is to explain the fundamental concepts
of artificial life and complex systems, and then to outline multi-agent simula-
tion and bo ttom-up simulation principles.
The significance of bottom-up simulation o f complex systems is analyzed
in the summary of this chapter. The simulations presented in this book take
a “constructive approach.” This is an engineering approach that deepens un-
derstanding by building fundamental models and then observing the behavior
of entire models built with those fundamental building blocks. The goal is to
“create and understand” complex systems and artificial life.
The complex systems covered here are limited to a c tual sy stems in the
real world becaus e the objective of the constructive approach is to under-
stand actual phenomena. Research is carried out sequentially in five stages.
It gradually shifts from an abstract level to more concrete levels, and insights
obtained in previous stages will be applied in subsequent stages. The research
will go back to stage 1 after the completion of s tage 5, and this procedure is
repeated to increase understanding of the s ubject.
Stage 1 Make something, which may not necessa rily correspon d to the subject,
that mimics the behavior of the subject. For example, models may be
simulated as multi-agent simulations of complex systems such as in
Swarm (Chapter 3).
Stage 2 Make something that qualitatively mimics the behavior of the subject.
An example is the reproduction of ant ma rches (Section 5.1).
Stage 3 Make something that q uantitatively mimics the behavior of the subject.
Appropriate adjustment of the parameters of the simulator and compar-
isons with observable examples are necessary at this stage.
249
250 Agent-Based Modeling and Simulation with Swarm
Stage 4 Hypotheses are tested, and the behavior o f the model is adjusted to
correspond to the actual behavior of the subject in the rea l world. Hy-
potheses are built to e xplain phenomena in natural sc ience . Experiments
are designed to verify whether the hypotheses hold. Similarly, the re sults
of simulations are compared with the results of experiments in the real
world as “hypothesis verification,” aiming to connect the model to the
real world.
Stage 5 Understand the origin of behavior in the subject, relate to the real world,
and explain the cause and effect based on actual physical and chemical
mechanisms. The goal is to understand not only the behavior of the
subject but also to understand the underlying factors and th e effects that
the behavior c auses. In biological terms, the objective is to understand
proximate (physiological) and final (biological or evolutional) causes.
There has been much discussion on the guiding methodology on how to
carry out research in the field of artificial life (see Section 4.4.1). For example,
Barandiaran and More no categorized artificial life into following four mod-
els [9].
Generic model Aims to deduce generic properties that e xist in any complex
system.
Conceptual model Aims to formulate and understand concepts such as evo-
lution or emergence.
Functional model Aims to understand specific sys tems with emergent prop-
erties.
Mechanistic model Aims to realistically reproduce the behavior of the sub-
ject model.
The relation between this classification of artificial life and the constructive
approach is shown in Fig. 8.1. The complete correspondence to the subject in
stage 5 is close to the mechanistic model. However, while the goal of a mech-
anistic model is to relate to the behavior of the s ubject, our approach aims
to relate to factors that cause the behavior of the subject. Functional models
model existing systems and corres pond to stage 4 because the focus is mainly
on emergence and evolution. In contrast, generic and co nce ptual models do not
model actual subjects; therefore, strictly there is no one-to-o ne relationship
with the stages in the constructive approach. However, these models can be
considered as components of stages 2 to 4 because of the common objective of
qualitatively a nd quantitatively clarifying the properties of co mplex systems.
Each model in artificial life rese arch has a different viewpoint. On the
other hand, the constructive a pproach progresses with research in steps with
different viewpoints to understand existing phenomena. Deeper insight is ob-
tained by connecting to the actual world, and then research is sta rted from
Conclusion 251
䌓䌴䌡䌧䌥㩷 䋱㩷
䌓䌴䌡䌧䌥㩷 㪉㩷
䌓䌴䌡䌧䌥㩷 㪊㩷
䌓䌴䌡䌧䌥㩷 㪋㩷
䌓䌴䌡䌧䌥㩷 㪌㩷
Functional model
Mechanistic model
Generic
model
Conceptual
model
㪘㫉㫋㫀㪽㫀㪺㫀㪸㫃㩷㫃㫀㪽㪼㩷㪸㫇㫇㫉㫆㪸㪺㪿㩷
㪚㫆㫅㫊㫋㫉㫌㪺㫋㫀㫍㪼㩷㪸㫇㫇㫉㫆㪸㪺㪿㩷
FIGURE 8.1: The classification of artificial life and the co nstructive ap-
proach.
stage 1 again using this insight. Repeating this appr oach will allow deeper
understanding of the subject.
Tables 8.1 and 8.2 show the relations between related research and the
constructive approach. Below we explain how this research corresponds to the
stages.
Related research 1: Pattern formation
Alan Turing proposed that morphogenesis ca n be modeled by the reactio n–
diffusion of morphogens, which are virtua l chemicals (see Section 7.7 for de-
tails). Genes corr e sp onding to morphogens have been fo und recently, and re-
search is under way. For instance, Yamaguchi e t al. are investigating how pat-
terns found on fish skin form at the genetic level based on Turing’s rea c tion–
diffusion model [129]. This work first built hypo these s based on the behavior
of an ac tual model, and then s imulations were carried out to finally derive the
relationship to the model at the experimental biology level. Hence, this is an
example where the constructive a pproach is taken.
Takeuchi et al. expanded the Turing model into a model that can explain
the formation of bacteria and cancer cell colonies [112]. By controlling pa-
rameters, they verified tha t sufficient nutrition conditions would r e sult in a
spherical cancer tissue w hereas deficient nutrition results in slow growth and
an irregular, complicated colony.
Synthetic biology is a promising new field of genetic engineering in this
direction [7, 111]. In synthetic biology, mathematical models are us e d to test
biological hypotheses and observations, and to predict the possible behaviors
of a desig ned gene circuit [49, 92]. These models serve as blueprints for novel
synthetic biological systems, making the engineering of biolog y easier and
more reliable. Evolutionary computation is a key tool in this field [58, ch. 6.5].
Related research 2: Artificial societies and artificial markets
252 Agent-Based Modeling and Simulation with Swarm
TABLE 8.1: Relations between re lated research a nd the c onstructive approach (1).
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5
Pattern
formation
Reproduction in a simulator [this book , Chapter
7]
Appropriate adjust-
ment of parameters
for pattern formation,
hypothesis verification
on reactio n–diffusion
model for patterns
of fish [75], colony
formation of skin
cancer [61]
Genetic-level understand-
ing of r e action–diffusion
model for patterns of
fish [75], understanding
concrete cr iteria for choos-
ing males and its stim-
ulus process [129]. Syn-
thetic biology approach [7,
49, 9 2, 111]
Artificial
societies
and ar-
tificial
markets
Reproduction in a simulator [this book , Chapter
6]
Confirming s imila rities
in wars and dealing
[this book, Sections 6.5
and 6.7], verification of
hypotheses such as the
bandwagon effect [62]
Understanding emergence
phenomena at the mi-
croscopic level (freq uency
distribution of rate fluc-
tuation, contrary opinion
phenomena) through in-
terviews with dealers [62]
Acoustic
conso-
nance and
dissonance
perception
model
Theory of beats
(Helmholtz),
dissonance per-
ception model
(Kameoka
and Kuriya-
gawa) [63, 64]
Investigating
similarity to
consonance
and dissonance
perceived by
humans [63, 64]
Agreement
with psycho-
logical experi-
ments [63, 64]
Prediction and verifica-
tion of consonance in
arbitrary sound [63, 64]
Frequency perception
mechanism of the inner
ear and relation to model
Conclusion 253
TABLE 8 .2: Relations between re lated research and the constructive approach (2).
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5
Biological
sp e c iation
Reproduction in a simulator [17, 84] Inve stigating
the number of
sp e c ies that will
emerge [1 7, 84]
Investigation
of how muta-
tion influences
sp e c iation [84]
Agreement with ex-
periments on bacte-
ria [84]
Foraging of
animals
Simulation using a classifier system [56] Acquisition of op-
timized foraging
rules [56]
Foraging based
on a numeri-
cal model [26],
mimicry and
search [94]
Exper iment to verify
model on cognition
during foraging [36]
Evolution-
ary robotics
Evolving virtual creatures [107], building
robots that ca n perform specific tasks [11,
27, 39]
GOLEM project
(NASA) [96],
morphogenesis
of robots [117],
building robots
capable of au-
tonomous learn-
ing [11, 27, 39]
Exper iment with walking humanoids
with brains mimicking monkeys [68],
exp e riment with robots that control in-
sects [113], building robots that can
perform specific tasks [11, 27, 39]
Emergence
in army ants
Reproduction in a
simulator [this book,
army ant]
Reproduction of ap-
propriate shortcuts
[this book, army ant]
Observation of
collective deter-
mination [this
book, Section 2.5]
Hypothesis
verification of
collective deter-
mination [this
book, Section 2.6]
Connection with ac-
tual army ants
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.