Theory in Biosciences manuscript No.

(will be inserted by the editor)

Guiding the Self-organization of Random Boolean Networks

Carlos Gershenson

Received: date / Accepted: date

Abstract Random Boolean networks (RBNs) are models of genetic regulatory networks. It is useful to
describe RBNs as self-organizing systems to study how changes in the nodes and connections affect the
global network dynamics. This article reviews seven different methods for guiding the self-organization of
RBNs. In particular, the article is focussed on guiding RBNs towards the critical dynamical regime, which
is near the phase transition between the ordered and dynamical phases. The properties and advantages
of the critical regime for life, computation, adaptability, evolvability, and robustness are revised. The
guidance methods of RBNs can be used for engineering systems with the features of the critical regime,
as well as for studying how natural selection evolved living systems, which are also critical.

Keywords guided self-organization · random Boolean networks · phase transitions · criticality ·

adaptability · evolvability · robustness

1 Self-organization and how to guide it

The concept of self-organization originated within cybernetics (Ashby, 1947; von Foerster, 1960; Ashby,
1962) and has propagated into almost all disciplines (Nicolis and Prigogine, 1977; Camazine et al,
2003; Skår and Coveney, 2003). Given the broad domains where self-organization can be described, its
formal definition is problematic (Gershenson and Heylighen, 2003; Heylighen, 2003; Gershenson, 2007;
Prokopenko et al, 2009). Nevertheless, we can use the concept to study a wide variety of phenomena.
To better understand self-organization, the following notion can be used: A system described as self-
organizing is one in which elements interact in order to achieve dynamically a global function or behavior
(Gershenson, 2007, p. 32). In other words, a global pattern is produced from local interactions.
Examples of self-organizing systems include a cell (molecules interact to produce life), a brain (neu-
rons interact to produce cognition), an insect colony (insects interact to perform collective tasks), flocks,
schools, herds (animals interact to coordinate collective behavior), a market (agents interact to define
prices), traffic (vehicles interact to determine flow patterns), an ecosystem (species interact to achieve
ecological homeostasis), a society (members interact to define social properties such as language, culture,

This work was partially supported by SNI membership 47907 of CONACyT, Mexico.

Instituto de Investigaciones en Matemáticas Aplicadas y en Sistemas

Universidad Nacional Autónoma de México
Ciudad Universitaria, A.P. 20-726
01000 México D.F. México
E-mail: [email protected] http://turing.iimas.unam.mx/~cgg
2

fashion, esthetics, ethics, and politics). In principle, almost any system can be described as self-organizing
(Ashby, 1962; Gershenson and Heylighen, 2003). If a system has a set of “preferred” states, i.e. attrac-
tors, and we call those states organized, the system will self-organize towards them. It is useless to enter
an ontological discussion on self-organization. Rather, the question is: when is it useful to describe a
system as self-organizing? For example, a cell can be described as self-organizing, but also as a Boolean
variable (0=dead, 1=alive). Which description is more accurate? It depends on the aim of the description
(model). A model cannot be judged independently of the context where it is used.
Self-organization is a useful description when at least two levels of description are present (e.g.
molecules and cells, insects and colony) and we are interested in studying the relationship between the
descriptions at these two levels (scales). In this way, one can describe how the interactions at the lower
level affect the properties at the higher level. When only the interactions at the lower level are defined,
the system can adapt to novel situations and be robust to perturbations, since the precise global behavior
is not predefined. Because of this, the properties of self-organizing systems can be exploited in design
and engineering (Gershenson, 2007; Watson et al, 2010).
The balance between self-organization and design is precisely the aim of guided self-organization
(GSO) (Prokopenko, 2009). Although it is difficult to define, GSO can be described as the steering of the
self-organizing dynamics of a system towards a desired configuration. Cybernetics (Wiener, 1948; Ashby,
1956) had already a similar aim, although there was a stronger focus on control and communication, as
opposed to self-organization and information.
The dynamics of self-organizing systems lead them to an “organized” state or configuration. However,
there can be several potential configurations available. The emerging study of GSO explores the con-
straints and conditions where self-organizing dynamics can be lead to a particular configuration. Similar
to several “synthetic” approaches (Steels, 1993), GSO can be useful on the one hand for understanding
how natural systems achieve GSO and on the other hand for building artificial systems capable of GSO.
This paper focuses on both aspects, exploiting the generality of random Boolean networks (RBNs):
First, how can evolution guide the self-organization of genetic regulatory networks? Second, how can we
manipulate RBNs to guide their self-organization towards a desired regime?
In the next section, random Boolean networks are briefly reviewed. In Section 3 the self-organization
of RBNs is described. Section 4 mentions seven different ways in which the self-organization of RBNs
can be guided. Section 5 presents a discussion. Conclusions close the paper.

2 Random Boolean networks

Random Boolean networks (RBNs) were originally proposed as models of genetic regulatory networks
(Kauffman, 1969, 1993). However, their generality has triggered an interest on them beyond their original
purpose (Aldana-González et al, 2003; Gershenson, 2004).
A RBN consists of N nodes linked by K connections each. Nodes are Boolean, i.e. their state is
either “on” (1) or “off” (0). The state of a node at time t + 1 depends on the states of its K inputs at
time t by means of a Boolean function. Connections and functions are chosen randomly when the RBN
is generated and remain fixed during its temporal evolution. The randomly generated Boolean functions
can be represented as lookup tables that represent all possible 2K combinations of input states. Fig. 1
shows an example of a part of a RBN, where every node has exactly two inputs, i.e. K = 2. Table 1
shows an arbitrary lookup table to update the state of one of the nodes. The dynamics of a RBN with
N = 40, K = 2 can be seen in Fig. 2.
Since RBNs are finite (they have 2N possible states) and deterministic, eventually a state will be
revisited. Then the network will have reached an attractor. The number of states in an attractor determine
the period of the attractor. Point attractors have period one (a single state), while cyclic attractors have
periods greater than one (multiple states, e.g. four in Fig. 2). A RBN can have one or more attractors.
The set of states visited until an attractor is reached is called a transient. The set of states leading to an
attractor form its basin. The basins of different attractors divide the state space. RBNs are dissipative,
 3

... n o p q r s ...

Fig. 1 Example of a RBN with connectivity K = 2, i.e. the state of nodes is determined by the state of two
other nodes. Note that all nodes have two inputs, but not necessarily two outputs, e.g. node n affects four other
nodes, while node o does not affect any other node.

Table 1 Lookup table to update node o depending on the state of nodes n and p. Lookup tables include all
possible combinations of inputs, i.e. 2K rows. Different nodes will have different lookup tables, i.e. Boolean
functions.

n(t) p(t) o(t+1)

0 0 1
0 1 0
1 0 0
1 1 1

Fig. 2 Temporal evolution of a RBN with N = 40, K = 2 for a random initial state. Dark squares represent 0
and light squares represent 1. Time flows to the right, i.e. columns represent states of the network at a particular
time, while a row represents the temporal evolution of the state of a node. Taken from RBNLab (Gershenson,
2005).

i.e. many states can flow into a single state (one state can have several predecessors), but from one
state the transition is deterministic towards a single state (one state can have only one successor). the
number of predecessors is also called in-degree. States without a predecessor are called “Garden of Eden”
(GoE) states (in-degree=0), since they can only be reached from an initial condition. Fig. 3 illustrates
the concepts presented above.
Note the difference between the topological network of a RBN (e.g. Fig. 1)—which represents how the
states of nodes affect each other—and its state network (e.g. Fig. 3)—which represents the transitions of
the whole state space. In the state network, each node represents a state of the network, i.e. there are 2N
nodes in the state network, while RBN nodes are represented in the topological network, i.e. there are N
nodes in the topological network. One of the main topics of RBN research is to understand how changes
in the topological network (lower scale) affect the state network (dynamics of higher scale), which is far
from obvious.
RBNs are a type of discrete dynamical network (Wuensche, 1998), i.e. space, time, and states are
discrete. RBNs are generalizations of Boolean cellular automata (von Neumann, 1966; Wolfram, 1986,
4

...
G
A
...

H B

Fig. 3 Example of state transtions. B is a successor state of A and a predecessor of C. States can have many
predecessors (e.g. B), but only one successor. G is a Garden of Eden state since it has no predecessors. The
attractor C → D → E → F → C has a period four.

2002), where the states of cells are determined by K neighbors, i.e. not chosen randomly, and all cells
are updated using the same Boolean function (Gershenson, 2002).

3 Self-organization in random Boolean networks

Random Boolean networks can be described as self-organizing systems simply because they have at-
tractors. If we describe the the attractors as “organized”, then the dynamics self-organize towards them
(Ashby, 1962). Still, a better argument in favor of this description is that we are interested in under-
standing how the interactions between nodes (lower scale) affect the network dynamics and properties
(higher scale). The concept of self-organization allows us to describe and relate both scales and their
interactions under the same framework.
The self-organization of RBNs can also be interpreted in terms of complexity reduction. For example,
the human genome has approximately 30,000 genes. Thus, in principle, each cell could be in one of the
230,000 possible states of that network. This is much more than the estimated number of elementary
particles arising from the Big Bang. However, there are only about 300 cell types (attractors (Kauffman,
1993; Huang and Ingber, 2000)), i.e. cells self-organize towards a very limited fraction of all possible
states. The main question addressed by these review paper is: in which ways can the self-organization of
random Boolean networks be guided?
Before presenting multiple answers to that question, it is convenient to understand the different
dynamical behaviors that RBNs can have (Wuensche, 1998; Gershenson, 2004). There are two dynamical
phases: ordered and chaotic. The phase transition is characterized by its criticality and is also known as
the “edge of chaos” (Kauffman, 1993).
In the ordered phase, most nodes do not change their state, i.e. they are static. RBNs are robust
in this phase, i.e. damage does not spread through the network, since most nodes do not change. Also,
similar states tend to converge to the same attractor. On average, states have many predecessors, which
leads to a high convergence (many states go to few states), short transient times, and a high density of
Garden of Eden states, i.e. the percentage of all states without a predecessor is high.
In the chaotic phase, most nodes are changing their state. Thus, damage spreads through chaotic
networks. Thus, RBNs are fragile in this phase. Similar states tend to diverge towards different attractors.
On average, states have few predecessors, which leads to a low convergence, very long average transient
times, and a relatively lower density of Garden of Eden states.
 5

In the critical regime, i.e. close to the transition between the ordered and chaotic phases, the extremes
of both phases are balanced: some nodes change and some are static. Therefore, damage or changes
can spread, but not necessarily through all the network. Similar states tend to lie in trajectories that
neither converge nor diverge in state space (Kauffman, 2000, p. 171). Few nodes have many predecessors,
while many nodes have few predecessors. Actually, the in-degree distribution approximates a power law
(Wuensche, 1998). There is medium convergence. It has recently been found that RBNs near the critical
regime maximize information storage and coherent information transfer (Lizier et al, 2008), as well as
maximize Fisher information (Wang et al, 2010).
It has been argued that computation and life necessarily lie at the edge of chaos (Langton, 1990;
Kauffman, 1993, 2000). Moreover, there is experimental evidence that the genetic networks of organisms
from at least four different kingdoms are near or within the critical regime (Balleza et al, 2008). The
tendency towards criticality can be explained because of the following: On the one hand, ordered dy-
namics produce stability (robustness) which is desirable for preserving information (memory). However,
a static system is not able to compute or adapt. On the other hand, chaotic dynamics give variability
(exploration), which is necessary for computation and adaptation. A balance is reached in the critical
regime, where the advantages of both phases can coexist: there can be enough stability and robustness
to preserve information and enough variability to compute and explore. For this reason, it becomes a
relevant question to ask how can we guide the self-organization of RBNs towards the critical regime.
Being general models, the answers will give us information on how to achieve the same guidance within
particular systems.

4 Guiding the self-organization of random Boolean networks

The criticality of RBNs can depend on many different factors. These factors can be exploited—by
engineers or by natural selection—to guide the self-organization of RBNs towards the critical regime.

4.1 p

One of the most obvious factors affecting the dynamics of a RBN is the probability p of having ones on
the last column of lookup tables (Derrida and Pomeau, 1986). If p = 1, then all values in lookup tables
will be one, so actually there will be no dynamics: all nodes will have a state of one after one iteration,
independently on the initial state. The same case but with zero occurs for p = 0. When p = 0.5 there is
the maximum variability possible in the lookup tables, i.e. no bias. As p approaches 0 or 1, RBNs tend
to be more static, i.e. in the ordered regime.

4.2 K

One of the most important factors determining the dynamical phase of RBNs is the connectivity K
(Derrida and Pomeau, 1986; Luque and Solé, 2000). For p = 0.5, the ordered phase is found when
K < 2, the chaotic phase occurs for K > 2, while the critical regime lies at K = 2. As p tends towards
one or zero, the transition moves towards greater values of K.
If we focus on a single node i and calculate the probability that damaging its state (be it 0 or 1)
will percolate changes through the network, then it is clear that the probability will increase with the
connectivity K. We can choose a node j from one of the nodes that i can affect. There is a probability p
that j will be 1, and thus a damage in i will modify j with a probability 1 − p. The complementary case
is the same. Now, for K nodes, we can expect that at least one change will occur if hKi 2p(1 − p) ≥ 1
(Luque and Solé, 1997), i.e. chaos. Generalizing, the critical connectivity becomes (Derrida and Pomeau,
1986):
6

1
hKi = (1)
2p(1 − p)

4.3 Canalizing functions

A canalizing function (Kauffman, 1969; Stauffer, 1987; Szejka and Drossel, 2007) is one in which at
least one of the inputs has one value that is able to determine the value of the output of the function,
regardless of the other inputs (Shmulevich and Kauffman, 2004). In other words, the non-canalizing
inputs of canalizing functions are not relevant. For this reason, if there is a bias favoring canalizing
functions, the phase transition will move towards greater values of K. This is possible because in practice
non-canalizing inputs are “ficticious”, i.e. removing them does not affect the state space nor the dynamics
of a RBN.

4.4 Topology

Until recently, RBN studies used either a homogeneous topology (K is the same for all nodes) or a normal
topology (there is an average hKi inputs per node). However, the topology can have drastic effects on the
properties of RBNs. On the one hand, topologies with more uniform rank distributions, as those used
commonly, exhibit more and longer attractors, but with less correlation in their expression patterns.
On the other hand, skewed topologies exhibit less and shorter attractors, but with more correlations
(entropy and mutual information) (Oosawa and Savageau, 2002). A balance between these two extremes
is achieved with scale-free topologies, i.e. few nodes have many inputs, while many nodes have few inputs.
This advantageous balance can be used to explain why most natural networks have a scale-free topology
(Aldana and Cluzel, 2003).
RBNs with a scale-free topology (Aldana, 2003) have been found to expand the advantages of the
critical regime into the ordered phase, since well-connected elements can lead to the propagation of
changes, i.e. adaptability even when the average connectivity would imply a static regime. It can be said
that a scale-free topology expands the range of the critical regime.

4.5 Modularity

It is well know that modularity is a prevalent property of natural systems (Callebaut and Rasskin-
Gutman, 2005) and a desired feature of artificial systems (Simon, 1996). Modules offer a level of organi-
zation that promotes at the same time robustness and evolvability (Wagner, 2005b). On the one hand,
damage within one module usually does not propagate through the whole system (robustness). On the
other hand, useful changes can be exploited to find new configurations without affecting the functionality
of other modules.
In the context of RBNs, initial explorations suggest that modules broaden the range of the critical
regime (Poblanno-Balp and Gershenson, 2010), i.e. a modular structure promotes critical dynamics.

4.6 Redundancy

Redundancy consists on having more than one copy of an element type. Duplication combined with
mutation is a usual mechanism for the creation of genes in eukaryotes (Fernández and Solé, 2004). When
there are several copies of an element type, changes or damage can occur to one element while others
continue to function.
 7

For RBNs (Gershenson et al, 2006), redundancy of links is not useful, since redundant links are
fictitious inputs, i.e. they do not affect the state space. However, a redundancy of nodes prevents muta-
tions from propagating through the network. Thus, redundant nodes increase neutrality (Kimura, 1983;
van Nimwegen et al, 1999; Munteanu and Solé, 2008), i.e. can “smoothen” rough landscapes. This is an
advantage for robustness and for evolvability, for the same reasons as the ones discussed concerning mod-
ularity, even when redundancy is a different mechanism from modularity, although both can potentially
be combined.

4.7 Degeneracy

Degeneracy—also known as distributed robustness—is defined as the ability of elements that are struc-
turally different to perform the same function or yield the same output (Edelman and Gally, 2001;
Fernández and Solé, 2004). As modularity, it also widespread in biological systems and a promotor of
robustness and evolvability. There is evidence that in genetic networks distributed robustness is equally
or more important for mutational robustness than gene redundancy (Wagner, 2005a,b).
To date, particular studies of degeneracy on RBNs are lacking. Nevertheless, it could be speculated
that degeneracy should promote critical dynamics, even when this still remains to be explored.

5 Discussion

In the previous section, a non-exhaustive list of mechanisms to guide the self-organization of RBNs
towards the critical regime was presented. Two categories of methods can be identified for guiding the
self-organization towards criticality: moving the phase transition (with p, K, or canalizing functions) or
broadening the critical regime (with a scale-free topology, modularity, redundancy, or degeneracy).
It can be speculated that natural selection can exploit these and probably other methods to guide the
self-organization of genetic regulatory networks towards the critical regime. There is evidence that some
of these methods are exploited by natural selection, but further studies are required to understand better
the mechanisms, their constraints, how they are related, and which ones have been actually employed
and to what degree by natural selection. In a similar way, engineers can guide the self-organization of
RBNs and related systems using these and other methods.
Concerning the methods that move the phase transition, if a RBN is in the ordered phase, one or
several of the following can be done:
– Adjust p towards a value of 0.5. This will increase variety in lookup tables, and thus increase the
number of nodes that change. In other words, dynamics will be promoted.
– Increase the connectivity K. More connections also promote richer dynamics.
– Decrease the number of canalizing functions (if any). Canalizing functions imply that some connec-
tions play no role on the dynamics. If these connections are changed, i.e. they become functional,
dynamics will be richer.
If the RBN is in the chaotic phase, complementary measures can be taken:
– Adjust p farther from a value of 0.5. This will increase homogeneity in lookup tables, and less nodes
will be changing. This will decrease the damage sensitivity of the network, i.e. the dynamics will be
less chaotic and more stable.
– Decrease the connectivity K. Less connections promote stability.
– Increase the number of canalizing functions. Canalizing functions reduce the effect of having several
inputs, i.e. a high connectivity K, so changes cannot propagate as easily as with no canalizing
functions.
Concerning the methods that broaden the critical regime, even when they are different, all of them
can be exploited to guide the self-organization of RBNs towards the critical regime:
8

– Promote a scale-free topology. When few nodes have many connections and most nodes have few
connections, the desirable properties of the critical regime extend beyond the extremely narrow
space of all possible networks that lies precisely on the phase transition (e.g. K = 2, p = 0.5).
This “criticality enhancement” is possible because most nodes are stable, but the nodes with several
connections (hubs) can trigger rich dynamics. In this way, changes can propagate through the RBN
in a constrained fashion.
– Promote modularity. Modules make it difficult for damages to spread through all the network, even
if the local connectivity (within a module) is high. In this way, chaotic dynamics can be constrained
within modules. This prevents avalanches, where changes on one node might cause drastic changes
in a large part of the network. Still, modularity allows for information flow between modules.
– Promote redundancy. Having more than one copy of a node (or module) implies that a change on
that node (or module) will not propagate through the network, since the redundant element(s) can
perform the same function. Apart from smoothening rough fitness landscapes (Gershenson et al,
2006), it has been noted that redundancy is a useful feature in evolvable hardware (Thompson,
1998).
– Promote degeneracy. The effect of degeneracy is similar to that of redundacy, but acting at a func-
tional level. Different components of a system perform the same function. In certain conditions,
degeneracy might be advantageous over redundancy, e.g. when a change affects all copies of the same
node (or module). Nevertheless, redundancy seems to be useful for exploration via duplication and
mutation.

Another way of guiding the self-organization of RBNs towards the critical regime would be to promote
certain properties as a part of the fitness function of an evolutionary algorithm. For example, one
could evolve critical RBNs trying to maximize input entropy variance (Wuensche, 1999), information
storage, information transfer (Lizier et al, 2008), and/or Fisher information (Wang et al, 2010). Another
criterium could be to try to approximate Lyapunov exponents to zero (Luque and Solé, 2000). All of
these properties characterize the critical regime. Thus, they can be used as a guidance of the evolutionary
search. Nevertheless, it should be noted that guiding the self-organization of RBNs with a fitness function
that promotes criticality is not useful to explain how natural systems evolved their criticality. Still, they
are a valid approach for engineering critical systems.

5.1 Why criticality?

Some of the advantages of the critical regime were already presented, namely the balance between
stability and variance that are requirements for life (Kauffman, 1993, 2000) and computation (Langton,
1990). In addition, the critical regime is also advantageous for adaptability, evolvability, and robustness.
Adaptability can be understood as the ability of a system to produce advantageous changes in response
to a state of its environment that will help the system to fulll its goals (Gershenson, 2007). Suppose that a
system that is modelled by a RBN (such as a genetic regulatory network) is situated in an unpredictable
environment. It is desirable that the system will be able to adapt to changes in the environment. Does
criticality increase adaptability? Not by itself, but it is useful. An ordered RBN will not be able to adapt
so easily, because most changes will have no effect on the dynamics, so there will be no response to the
environmental change. A chaotic RBN will pose the opposite difficulty: most changes will have a strong
effect on the dynamics. Thus, it is highly probable that some of the functionality of the system will be
lost. A balance is achieved by a critical RBN: changes can have an effect on the dynamics, but their
propagation can be constrained, preserving most of the functionality of the system.
Evolvability is the ability of random variations to sometimes produce improvement (Wagner and
Altenberg, 1996). It can be seen as a particular type of adaptability, where changes occur from genera-
tion to generation. RBN evolution has already been explored (Stern, 1999; Lemke et al, 2001). For the
same reasons as those exposed for adaptability, evolvability will be higher for critical RBNs. Within an
evolutionary context, however, it is also important to mention the advantages of criticality to scalability
 9

(Simon, 1996), i.e. the ability to acquire novel functionalities. Since ordered RBNs have restricted dynam-
ics and interactions, they cannot integrate novel functions too easily. Chaotic RBNs are also problematic,
since novel functions most probably change the existing functionality. Critical RBNs are scalable, since
novel functions can be integrated without altering existing functionalities. Intuitively, modularity pro-
motes scalability in the most straightforward way, although other methods—i.e. critical RBNs without
modularity—also allow for scalability.
A system is robust if it continues to function in the face of perturbations (Wagner, 2005b; Jen, 2005).
Robustness is a desirable property to complement adaptability and evolvability, since changes in the
environment (perturbations) can damage or destroy a system before it can adapt or evolve. It is clear
that evolution is not possible without robustness. Chaotic RBNs are not robust, since small perturbations
produce drastic changes. Ordered RBNs are very robust, since they can resist most perturbations without
producing changes. And when changes are produced, these do not propagate. However, ordered RBNs do
not offer rich dynamics. Critical RBNs offer both advantages: rich dynamics and robustness (although
not as high as the one of ordered RBNs. Note that the most robust RBNs are those without dynamics,
e.g. with p = 1.).
Topology and modularity seem to be more relevant for evolvability, while redundancy and degeneracy
seem to be more relevant for robustness. However, evolvability and robustness are interrelated properties
(Yu and Miller, 2001; Ebner et al, 2002; Wagner, 2005b), both of them desirable in natural and artificial
systems.

6 Conclusions

This paper described random Boolean networks (RBNs) as self-organizing systems. Given the advantages
of the critical regime of RBNs, different methods to guide the self-organization of RBNs towards criticality
were reviewed.
One can ask: which comes first, criticality or the methods that promote it? Do they always come
hand in hand? It seems not, since criticality can be present without canalizing functions, modularity,
redundancy, degeneracy, or scale free topologies. However, these properties facilitate (guide) criticality.
Therefore, there is a selective pressure in favor of these properties. Which are the pressures that have
actually guided genetic regulatory networks towards criticality (Balleza et al, 2008) is an open question.
Which methods of the ones presented have actually been exploited by natural selection is another open
question, although there is evidence of several of them at play. Yet another open question is how are
the different methods related. This question actually comprises a whole set of questions, e.g. how are
scale-free and modular topologies related? Is there an advantage of having both a scale-free topology
and modularity over only one of them? When is redundancy or degeneracy preferable? What are the
differences and advantages of critical RBNs produced with one or several of the presented methods? How
are different methods related to adaptability, evolvability, and robustness? What is the proper balance
between evolvability and robustness?
This long list of relevant questions, which could easily continue growing, should motivate researchers
to continue exploring RBNs, their self-organization, and methods for guiding it.

References

Aldana M (2003) Boolean dynamics of networks with scale-free topology. Physica D 185(1):45–66, URL
http://dx.doi.org/10.1016/S0167-2789(03)00174-X
Aldana M, Cluzel P (2003) A natural class of robust networks. Proceedings of the National Academy
of Sciences of the United States of America 100(15):8710–8714, DOI 10.1073/pnas.1536783100,
URL http://www.pnas.org/content/100/15/8710.abstract, http://www.pnas.org/content/100/
15/8710.full.pdf+html
10

Aldana-González M, Coppersmith S, Kadanoff LP (2003) Boolean dynamics with random couplings.

In: Kaplan E, Marsden JE, Sreenivasan KR (eds) Perspectives and Problems in Nonlinear Science. A
Celebratory Volume in Honor of Lawrence Sirovich, Springer Applied Mathematical Sciences Series,
URL http://www.fis.unam.mx/%7Emax/PAPERS/nkreview.pdf
Ashby WR (1947) Principles of the self-organizing dynamic system. Journal of General Psychology
37:125–128
Ashby WR (1956) An Introduction to Cybernetics. Chapman & Hall, London, URL http://pcp.vub.
ac.be/ASHBBOOK.html
Ashby WR (1962) Principles of the self-organizing system. In: Foerster HV, Zopf, Jr GW (eds) Principles
of Self-Organization, Pergamon, Oxford, pp 255–278
Balleza E, Alvarez-Buylla ER, Chaos A, Kauffman S, Shmulevich I, Aldana M (2008) Critical dynamics
in genetic regulatory networks: Examples from four kingdoms. PLoS ONE 3(6):e2456, DOI 10.1371/
journal.pone.0002456, URL http://dx.plos.org/10.1371%2Fjournal.pone.0002456
Callebaut W, Rasskin-Gutman D (2005) Modularity: Understanding the Development and Evolution of
Natural Complex Systems. MIT Press
Camazine S, Deneubourg JL, Franks NR, Sneyd J, Theraulaz G, Bonabeau E (2003) Self-Organization in
Biological Systems. Princeton University Press, URL http://www.pupress.princeton.edu/titles/
7104.html
Derrida B, Pomeau Y (1986) Random networks of automata: A simple annealed approximation. Europhys
Lett 1(2):45–49
Ebner M, Shackleton M, Shipman R (2002) How neutral networks influence evolvability. Complexity
7(2):19–33, DOI 10.1002/cplx.10021, URL http://dx.doi.org/10.1002/cplx.10021
Edelman GM, Gally JA (2001) Degeneracy and complexity in biological systems. Proceedings of
the National Academy of Sciences of the United States of America 98(24):13,763–13,768, DOI
10.1073/pnas.231499798, URL http://www.pnas.org/content/98/24/13763.abstract, http://www.
pnas.org/content/98/24/13763.full.pdf+html
Fernández P, Solé R (2004) The role of computation in complex regulatory networks. In: Koonin EV,
Wolf YI, Karev GP (eds) Power Laws, Scale-Free Networks and Genome Biology, Landes Bioscience,
URL http://arxiv.org/abs/q-bio.MN/0311012
Gershenson C (2002) Classification of random Boolean networks. In: Standish RK, Bedau MA, Abbass
HA (eds) Artificial Life VIII: Proceedings of the Eight International Conference on Artificial Life, MIT
Press, pp 1–8, URL http://alife8.alife.org/proceedings/sub67.pdf
Gershenson C (2004) Introduction to random Boolean networks. In: Bedau M, Husbands P, Hutton
T, Kumar S, Suzuki H (eds) Workshop and Tutorial Proceedings, Ninth International Conference on
the Simulation and Synthesis of Living Systems (ALife IX), Boston, MA, pp 160–173, URL http:
//uk.arxiv.org/abs/nlin.AO/0408006
Gershenson C (2005) RBNLab. Http://rbn.sourceforge.net
Gershenson C (2007) Design and Control of Self-organizing Systems. CopIt Arxives, Mexico, URL http:
//tinyurl.com/DCSOS2007, http://tinyurl.com/DCSOS2007
Gershenson C, Heylighen F (2003) When can we call a system self-organizing? In: Banzhaf W, Christaller
T, Dittrich P, Kim JT, Ziegler J (eds) Advances in Artificial Life, 7th European Conference, ECAL
2003 LNAI 2801, Springer, Berlin, pp 606–614, URL http://uk.arxiv.org/abs/nlin.AO/0303020
Gershenson C, Kauffman SA, Shmulevich I (2006) The role of redundancy in the robustness of random
Boolean networks. In: Rocha LM, Yaeger LS, Bedau MA, Floreano D, Goldstone RL, Vespignani A
(eds) Artificial Life X, Proceedings of the Tenth International Conference on the Simulation and Syn-
thesis of Living Systems., MIT Press, pp 35–42, URL http://uk.arxiv.org/abs/nlin.AO/0511018
Heylighen F (2003) The science of self-organization and adaptivity. In: Kiel LD (ed) The Encyclo-
pedia of Life Support Systems, EOLSS Publishers, Oxford, URL http://pcp.vub.ac.be/Papers/
EOLSS-Self-Organiz.pdf
Huang S, Ingber DE (2000) Shape-dependent control of cell growth, differentiation, and apoptosis:
Switching between attractors in cell regulatory networks. Exp Cell Res 261:91–103
 11

Jen E (ed) (2005) Robust Design: A Repertoire of Biological, Ecological, and Engineering Case Studies.
Santa Fe Institute Studies on the Sciences of Complexity, Oxford University Press, URL http://
tinyurl.com/swtlz
Kauffman SA (1969) Metabolic stability and epigenesis in randomly constructed genetic nets. Journal
of Theoretical Biology 22:437–467
Kauffman SA (1993) The Origins of Order. Oxford University Press
Kauffman SA (2000) Investigations. Oxford University Press
Kimura M (1983) The Neutral Theory of Molecular Evolution. Cambridge University Press, Cambridge
Langton C (1990) Computation at the edge of chaos: Phase transitions and emergent computation.
Physica D 42:12–37
Lemke N, Mombach JCM, Bodmann BEJ (2001) A numerical investigation of adaptation in popula-
tions of random Boolean networks. Physica A 301(1–4):589–600, URL http://dx.doi.org/10.1016/
S0378-4371(01)00372-7
Lizier J, Prokopenko M, Zomaya A (2008) The information dynamics of phase transitions in random
boolean networks. In: Bullock S, Noble J, Watson R, Bedau MA (eds) Artificial Life XI - Proceedings
of the Eleventh International Conference on the Simulation and Synthesis of Living Systems, MIT
Press, Cambridge, MA, USA, pp 374–381, URL http://tinyurl.com/3xzx9fr
Luque B, Solé RV (1997) Phase transitions in random networks: Simple analytic determination of critical
points. Physical Review E 55(1):257–260, URL http://tinyurl.com/y8pk9y
Luque B, Solé RV (2000) Lyapunov exponents in random Boolean networks. Physica A 284:33–45, URL
http://tinyurl.com/trnd4
Munteanu A, Solé RV (2008) Neutrality and robustness in evo-devo: Emergence of lateral inhibition.
PLoS Comput Biol 4(11):e1000,226, DOI 10.1371/journal.pcbi.1000226, URL http://dx.doi.org/
10.1371%2Fjournal.pcbi.1000226
Nicolis G, Prigogine I (1977) Self-Organization in Non-Equilibrium Systems: From Dissipative Structures
to Order Through Fluctuations. Wiley
van Nimwegen E, Crutchfield JP, Huynen M (1999) Neutral evolution of mutational robustness. Pro-
ceedings of the National Academy of Sciences of the United States of America 96(17):9716–9720,
URL http://www.pnas.org/content/96/17/9716.abstract, http://www.pnas.org/content/96/17/
9716.full.pdf+html
Oosawa C, Savageau MA (2002) Effects of alternative connectivity on behavior of randomly constructed
Boolean networks. Physica D 170:143–161
Poblanno-Balp R, Gershenson C (2010) Modular random Boolean networks. In: ALife XII Proceedings,
MIT Press, accepted
Prokopenko M (2009) Guided self-organization. HFSP Journal 3(5):287–289, DOI 10.2976/1.3233933,
URL http://dx.doi.org/10.2976/1.3233933
Prokopenko M, Boschetti F, Ryan A (2009) An information-theoretic primer on complexity, self-
organisation and emergence. Complexity 15(1):11 – 28, DOI 10.1002/cplx.20249, URL http://dx.
doi.org/10.1002/cplx.20249
Shmulevich I, Kauffman SA (2004) Activities and sensitivities in boolean network models. Phys
Rev Lett 93(4):048,701, DOI 10.1103/PhysRevLett.93.048701, URL http://dx.doi.org/10.1103/
PhysRevLett.93.048701
Simon HA (1996) The Sciences of the Artificial, 3rd edn. MIT Press
Skår J, Coveney PV (eds) (2003) Self-Organization: The Quest for the Origin and Evolution of Structure,
Phil. Trans. R. Soc. Lond. A 361(1807), proceedings of the 2002 Nobel Symposium on self-organization
Stauffer D (1987) On forcing functions in Kauffman’s random Boolean networks. Journal of Statistical
Physics 46(3):789–794, DOI 10.1007/BF01013386, URL http://dx.doi.org/10.1007/BF01013386
Steels L (1993) Building agents out of autonomous behavior systems. In: Steels L, Brooks RA (eds) The
Artificial Life Route to Artificial Intelligence: Building Embodied Situated Agents, Lawrence Erlbaum
Stern MD (1999) Emergence of homeostasis and noise imprinting in an evolution model. PNAS 96:10,746–
10,751
12

Szejka A, Drossel B (2007) Evolution of canalizing boolean networks. EPJ B 56(4):373–380, DOI 10.
1140/epjb/e2007-00135-2, URL http://dx.doi.org/10.1140/epjb/e2007-00135-2
Thompson A (1998) Hardware Evolution: Automatic Design of Electronic Circuits in Reconfigurable
Hardware by Artificial Evolution. Distinguished dissertation series, Springer-Verlag
von Foerster H (1960) On self-organizing systems and their environments. In: Yovitts MC, Cameron S
(eds) Self-Organizing Systems, Pergamon, New York, pp 31–50
von Neumann J (1966) The Theory of Self-Reproducing Automata. University of Illinois Press, edited
by A. W. Burks
Wagner A (2005a) Distributed robustness versus redundancy as causes of mutational robustness. BioEs-
says 27(2):176–188, DOI 10.1002/bies.20170, URL http://dx.doi.org/10.1002/bies.20170
Wagner A (2005b) Robustness and Evolvability in Living Systems. Princeton University Press, Princeton,
NJ, URL http://www.pupress.princeton.edu/titles/8002.html
Wagner G, Altenberg L (1996) Complex adaptations and the evolution of evolvability. Evolution
50(3):967–976
Wang XR, Lizier J, Prokopenko M (2010) A fisher information study of phase transitions in random
boolean networks. In: Twelfth International Conference on the Simulation and Synthesis of Living
Systems (ALife XII), MIT Press, URL http://tinyurl.com/37qxgtn
Watson RA, Buckley CL, Mills R (2010) Optimisation in “self-modelling” complex adaptive systems.
Complexity URL http://eprints.ecs.soton.ac.uk/21051/, accepted
Wiener N (1948) Cybernetics; or, Control and Communication in the Animal and the Machine. Wiley
and Sons, New York
Wolfram S (1986) Theory and Application of Cellular Automata. World Scientific
Wolfram S (2002) A New Kind of Science. Wolfram Media, URL http://www.wolframscience.com/
thebook.html
Wuensche A (1998) Discrete dynamical networks and their attractor basins. In: Standish R, Henry B,
Watt S, Marks R, Stocker R, Green D, Keen S, Bossomaier T (eds) Complex Systems ’98, University
of New South Wales, Sydney, Australia, pp 3–21, URL http://tinyurl.com/y6xh35
Wuensche A (1999) Classifying cellular automata automatically: Finding gliders, filtering, and relating
space-time patterns, attractor basins, and the Z parameter. Complexity 4(3):47–66, URL http://
tinyurl.com/y7pss7
Yu T, Miller J (2001) Neutrality and the evolvability of boolean function landscape. In: Genetic
Programming, LNCS, vol 2038, Springer, pp 204–217, DOI 10.1007/3-540-45355-5 16, URL http:
//dx.doi.org/10.1007/3-540-45355-5_16