ArticlePDF Available

Abstract and Figures

FuturICT foundations are social science, complex systems science, and ICT. The main concerns and challenges in the science of complex systems in the context of FuturICT are laid out in this paper with special emphasis on the Complex Systems route to Social Sciences. This include complex systems having: many heterogeneous interacting parts; multiple scales; complicated transition laws; unexpected or unpredicted emergence; sensitive dependence on initial conditions; path-dependent dynamics; networked hierarchical connectivities; interaction of autonomous agents; self-organisation; non-equilibrium dynamics; combinatorial explosion; adaptivity to changing environments; co-evolving subsystems; ill-defined boundaries; and multilevel dynamics. In this context, science is seen as the process of abstracting the dynamics of systems from data. This presents many challenges including: data gathering by large-scale experiment, participatory sensing and social computation, managing huge distributed dynamic and heterogeneous databases; moving from data to dynamical models, going beyond correlations to cause-effect relationships, understanding the relationship between simple and comprehensive models with appropriate choices of variables, ensemble modeling and data assimilation, modeling systems of systems of systems with many levels between micro and macro; and formulating new approaches to prediction, forecasting, and risk, especially in systems that can reflect on and change their behaviour in response to predictions, and systems whose apparently predictable behaviour is disrupted by apparently unpredictable rare or extreme events. These challenges are part of the FuturICT agenda. Graphical abstract
Content may be subject to copyright.
arXiv:1204.4928v1 [nlin.AO] 22 Apr 2012
EPJ manuscript No.
(will be inserted by the editor)
Challenges in Complex Systems Science
Maxi San Miguel
1, a
, Jeffrey H. Johnson
, Janos Kertesz
, Kimmo Kaski
, Albert
, Robert S. MacKay
, Vittorio Loreto
, eter
, and Dirk
IFISC (CSIC-UIB), Campus Universitat Illes Balears, E-07071 Palma de Mallorca, Spain
Faculty of Mathematics, Computing & Technology, The Op en U niversity, MK7 6AA, UK
Institute of Physics, Budapest Univ. of Technology & Economics, Budafoki ´ut 8, Budapest
Dept. of Biomedical Engineering & Computational S cience, FI -00076 Aalto, Finland
Dept. Fisica Fonamental, Universitat de Barcelona, E-08028 Barcelona, Spain
Mathematics Institute & Centre for Complexity Science, University of Warwick, CV4 7AL
Sapienza Universty of Rome, Physics Dept., Rome, Italy
ISI Foundation, Turin, Italy
Institute for Particle and Nu clear Physics, Wigner Research Centre for Physics, Hungarian
Academy of Sciences, Budapest
Center for Complex Systems Studies, Kalamazoo College, Michigan, MI 49006, USA
ETH Z¨urich, Clausiusstrasse 50, 8092 Z¨urich, Switzerland
Abstract. FuturICT foundations are social science, complex systems
science, and ICT. The main concerns and challenges in the science of
complex systems in the context of FuturICT are laid out in this pa-
per with special emphasis on the Complex Systems route to Social Sci-
ences. This include complex systems having: many heterogeneous inter-
acting p arts; multiple scales; complicated transition laws; unexpected
or unpredicted emergence; sensitive dependence on initial conditions;
path-dependent dynamics; networked hierarchical connectivities; in-
teraction of autonomous agents; self-organisation; non-equilibrium dy-
namics; combinatorial ex plosion; adaptivity to changing environments;
co-evolving subsystems; ill-defined boundaries; and multilevel dy nam-
ics. In this context, science is seen as the process of ab stracting the
dynamics of systems from data. This presents many challenges includ-
ing: data gathering by large-scale experiment, p articipatory sensing
and social computation, managing huge distributed dynamic and het-
erogeneous databases; moving from data to dynamical models, going
beyond correlations to cause-effect relationships, understanding the re-
lationship between simple and comprehensive models with appropriate
choices of variables, ensemble modeling and data assimilation, mod-
eling systems of systems of sy stems with many levels between micro
and macro; and formulating new approaches to prediction, forecasting,
and risk, especially in systems th at can reflect on and change their
behaviour in response to predictions, and systems whose apparently
predictable behaviour is disrupted by apparently unpredictable rare or
extreme events. These challenges are part of the FuturICT agenda.
2 Will be inserted by the editor
1 Introduction
Simplicity and sparsity of scientific description have always been regarded as a the-
oretical virtue. Aristotle in the Posterior Analytics says: “We may a ssume the supe-
riority ceteris paribus of the demonstration which de rives from fewer postulates or
hypotheses.” Acco rding to Newton “Nature is ple ased with s implicity, and affects no t
the pomp o f s uperfluous causes”, while we learn from Einstein that “The grand aim
of all science ... is to cover the greatest possible number of empirical facts by logical
deductions from the smallest possible number of hypotheses or a xioms.”
Modern science started with the a nalysis of the simplest phenomena and physics,
a reductionist science par excellence, emerg ed as the leading example of how the
human mind can make sense of the apparent chaos of the phenomena surrounding
us. The key to the early success of physics was that it studied objects tha t could be
described in terms of a few va riables, could be well separated from their environment,
with well-targeted reproducible experiments that could be per formed on them. During
the course of its development physics has learned how to tackle problems that are
immensely more complicated than the free fall of balls from the Tower of Pisa, but the
reductionist program remains one of its core motivations. The dream of a theory of
everything” drives the quest for the ultimate building blocks of the Universe and for
the explanation of its origin - an endeavor c onstituting one of the frontiers of science.
However, as stated by P. W. Anderson in 197 2 [1 ] the reductionist hypothesis does
not by any means imply a “constructionist” hypothesis: the constructionist hypothesis
breaks down when confronted with the twin difficulties of sc ale and complexity. Most
of the objects of scientific inquiry shar e these difficulties. For example, a living being
cannot be describ ed in terms of a few variables, a human being cannot be separated
from the rest of society without altering its nature fundamentally, and the functional-
ity of our brain emer ges from the network of interacting neurons. These are examples
of what nowadays are called complex systems. A growing body of knowledge is being
accumulated about these complex systems, a large number of groups a re striving for
a deeper understanding of their common features and an ever richer set of concepts
and tools are being devised to tackle them. These developments are gradually leading
up to what we believe is becoming a coherent and fundamental science of complexity
[2,3]. Understanding the basic principles of complexity and emergent phenomena in
complex systems is the other frontier of present day science. If the goal of particle
physics is the ultimate analysis, that of complexity science is the ultimate synthesis.
To promote this synthesis is one of the main motivations of the FuturICT program.
Other motivation stems from the ever-increasing relevance of complex-systems ori-
ented approaches to social science [4]. I t may be surprising but the idea of a physical
modeling of social phenomena [5] is in some sense older than the idea of statistical
modeling of physical phenomena. The discovery of quantitative laws in the collective
properties of a large number of people, as revealed, for example, by birth and death
rates or crime statistics, was one of the catalysts in the development of statistics, a nd
it led many scientists and philo sophers to call for some quantitative understanding
of how such precise re gularities arise out of the apparently erratic behavior of single
individuals. Hobbes, Laplace, C omte, Stuart Mill, a nd many others shared, to a dif-
ferent extent, this line of thought [6]. This point of view was well known to Maxwell
and Boltzmann and probably played a role when they abandoned the idea of de scrib-
ing the trajectory of single particles and introduced a statistical description for gases,
laying the foundations of modern statistical physics. The value of statistical laws for
social s cience was foreseen also by Majorana [7]. But it is only in the past few years
that the idea of approaching society within the framework of statistical physics has
Will be inserted by the editor 3
transformed from a philosophical declaration of principles to a concrete research effort
involving a critical mass of scientists. The availability of new large databases as well
as the appearance of brand new social phenomena, mostly related to the Informatio n
and Communication Technologies, and the tendency of social scientists to move to-
ward the formulation of simplified models [8,9] and their q uantitative analysis [10],
have been instrumental in this change. Nowadays the understanding of the dynamics
of human societies and finding viable solutions to the enormous problems they are
facing is a matter not only of knowledge, but survival. Demogr aphic changes, migra-
tions, destruction of the environment, depletion of reso urces, the structural instability
of our economic and social systems are only s ome of the most prominent among these
problems. Futur IC T is devoted to the analysis and modeling of these complex and
interwoven processes.
2 Complex Systems
Conventional wisdom suggests that simple systems be have simply, complex behavior
arises from complex causes and that different systems behave differently. There is
ample evidence, even in the physical sciences, that these statements are unfounded
and not g enerally correct. The roots of such oversimplified views are close ly related to
the problems of complex systems themselves, e.g. the choice of appropriate variables,
the lack or multitude o f scales, multilevel structure in both space and time, the level
of description, and so on. It can be highly non-trivial to find the simple causes behind
complex behavior and to select appropriate variables by which the possible generic
behavior of the system would become apparent.
An important source of difficulties stems from the fundamental problem of the
level of detail and c omplexity needed for the understanding of the structure, func-
tion, and response of a complex system of interest. This is perhaps best gained with
the analysis of data from well-devised experiments or from various datasets and prop er
modeling. The success of this two-step empirical approach has to be judged in rela-
tion to the goal and purpose of the study process, while aiming for understanding,
, managing and even controlling the behavior of the system. In all this it
should be emphasized that modeling goes hand in ha nd with the availability and use
of the data. This is because the need for empirical data is paramount not only for
the understanding and exploratio n of the features and phenomena of the s ystems of
interest but also for calibrating and validating the models for impr oved usefulness in
predicting, forecasting, and managing the behavior of the s ystem.
Calibration can be a delicate ta sk and much attention should be paid to it. Given
the level of complexity and the related dimensionality of the pr oblem the amount
of data used for calibration should be chosen such that overfitting can be avoided
in order to exclude spurious dependencies. On the other hand - as follows from the
very nature of complex s ystems - a s much data as possible should be used. This issue
One of the challenges for complex systems science is to better understand the term
‘prediction’ and the part it plays in science and its applications. In this paper it will be used
generally to include various different ways for describing the future behaviour of systems.
Traditionally science has made point predictions that a system will be in a particular state
at a particular point in future time. Beyond relatively short horizons, point predictions of
systems that are sensitive to initial conditions become increasingly error prone, e.g. the
weather. This is so even when the underlying model is prefect, due to inevitable errors in
measuring initial conditions. Some predictions are probabilistic based on estimates of relative
frequencies, but this approach is inappropriate for ‘rare’ events of m easure zero. Predictions
of social systems may be self-contradictory as the system reflects on and changes behaviour
in response to the prediction, or self-fulfilling prophesies when policy forces events to happen.
4 Will be inserted by the editor
reflects the deep problem of computer scie nc e of finding the bala nc e between avoiding
both overfitting and oversimplification. Of course, it is also related to the problem of
finding the right variables mentioned previously.
In the new ICT-ba sed frame of Social Sciences, the main problem is often not data
availability but the challenge of extra cting releva nt knowledge from observational data
and in devis ing useful data acquisition for answering specific questions of the be havior
of the system of interest.
In relation to these specific questions we see an increasing role of more question-
driven research, where massive, ICT-based data are already collected with a specific
aim and e ven experiments are being devised (see Sect. 7.2). Such a trend could lead
to or open new frontiers in studies of ICT-related social systems. To summarize we
envisage that the study o f a complex system should proceed in the following steps in
which the use of data and models is essential:
i) Exploratio n, obse rvation and basic data acquisition,
ii) Identification of correlations, patterns, and mechanisms,
iii) Mode ling,
iv) Model validation, implementation and pre diction,
v) Construction of a theory.
In devising a model for a given system with a large number of constituents it
is useful first to discuss how one would characterize the basic constituents and/o r
governing laws (if known). In this context a distinction can be made between compli-
cated systems and complex systems. Complicated systems are viewed to have a large
number of components which behave in a well-understood way and have well-defined
roles leading to the resulting effect, e.g. mode rn airplanes with millions of physical
parts and even tens of millions of lines o f software. Complex sy stems typically have a
large number of components, w he re the interactions (however s imple they may be on
the individual level) lead to collective emergent behaviours that cannot, even qual-
itatively, be derived as a plain resultant from the individual components’ behavior.
Paramount examples of complex systems are our brain and our societies.
All domain-based sciences such as physics, chemistry, biology, psychology, sociol-
ogy, economics, robotics, medicine and business investigate systems that are complex
in one way or another. These sciences investigate their domains in depth, which con-
trasts with the emerging science of complex systems which intersects the domains
horizontally. By looking across the disciplines the methodology of complex systems
provides two new perspectives: the first is that apparently different systems may have
common pr operties and knowledge from one discipline can usefully feed into another;
the s e cond is that the scienc e of complex systems is trans-disciplinary and it is creat-
ing new methods to combine the dynamical theories of many interacting social and
technical subsystems. Unlike domain-based sciences such as those mentioned above,
complex systems science is integrative - a science of systems of systems across many
There is no agreement on what should b e the precise definition of complex and
there are many r e asons as to why a sy stem might be considered complex. Table 2
reports a list of features typical of complex systems along with concrete examples of
systems displaying those features. Of course many systems could exhibit several of
these features. Any one of them can make systems appear co mplex, but together they
can make systems very difficult to understand and control [11]. A key characteris tics
of complex systems is their ability to reconfigure themselves to create new systems
with completely different properties.
Complex sys tems such as cities, the human body, or economies have dynamics
at many differe nt scales. The presence of many s cales or, even worse, the conflu-
ence of scales and lack of a characteristic s c ale that would allow the breakdown of
Will be inserted by the editor 5
many heterogeneous interacting parts cities, companies, climate, crowds
political parties, ecosystems
complicated transition laws markets, disease transmission, cascading failure
rioting, professional training
unexpected or unp redictable emergence chemical systems, accidents, system breakdown
sp ontaneous social initiatives, foot and mouth disease
sensitive dependence on initial conditions weather systems, investments
traffic jams, forest fires
path-dependent dynamics the evolution of the qwerty keyboard,
racial conflicts, first to market
networked hierarchical connectivities social networks, ecosystems ,the Internet
voting systems, postal systems
interactions of autonomous agents road traffic, dinner parties
housing markets, soccer games, crowd dynamics
self-organisation or collective shifts revolutions, fashions, choirs
demonstrations, property rental markets
non-equilibrium dynamics fighter aircraft, share prices, the weather
armed conflict, social networking
combinatorial explosion chess, communications systems,
data states for a computer program
adaptivity to changing environments biological systems,manufacturing design
retail systems, rebranding
co-evolving subsy stems land-use, transp ortation
computer virus software
ill-defined boundaries genetically modified crops, nations,
pollution, terrorism, markets
multilevel dynamics companies, armies, governments
aircraft, Internet, transportation
Table 1. Reasons why systems might be considered to be complex
the problem into sub- problems makes a standard reductionist micro-macro approa ch
difficult. This leads to the appearance of fat tails and self-similar distributions, e.g.
Pareto-distributions of wealth, company size, capitalization, etc..
Socio-technical systems have strong interactions leading to collective behaviour,
building up macroscopic structures that act as top- down constraints on the micro-
scopic degrees of freedo m (mode slaving), e.g. long wavelength spin waves acting as an
external field o n the individual spins, or institutions, conventions, traditions, culture,
etc. acting on, and largely conditioning, the agents that cr eated them. Thus “the
whole is more (or less) than the sum of its parts”: e.g. cutting a horse in two does not
result in two small horses; the merger of two successful companies (or universities)
rarely cre ates a better co mpany (but may create a monopoly); and uniting disparate
nations into Yugoslavia led to disas ter 70 years later. Underlying the formation of
wholes is the emergence of strong, long-range interactions and correla tions in c om-
plex systems, that link distant parts. Complex systems are likely to feature non-local
interactions in space and time. This property often makes systems sensitive no t only
to initial conditions, but also to boundary conditions and small changes in the control
This is tightly related to “irr educibility”, i.e., the impossibility of describing a
complex system in terms of a few variables. (The local susce ptibility is the sum of
correlations measured from the given local element: if correlations are long ra nged
and the system is heterogeneous, the local susceptibility depends on a large number
6 Will be inserted by the editor
of variables.) Multi-attractor structure and the resulting path-dependence are related
aspects. Complex systems may sometimes evolve slowly, but they ar e never in equi-
librium (unless dead). For example: it is hard to predict biological evolution, but it
may be possible to rationalize backwards.
Beyond a certain level of complexity systems not only reflect their prior evolution,
but also start to learn, and modify their behaviour according to changes in the en-
vironment, conditions, etc. At even higher levels they self-reflect, r e act to what they
“think” about themselves, or what is thought about them, e.g. self- fulfilling prophe-
cies, collective myths, etc.. For example, tourism may cause prices to rise as traders
see the opportunities, but ma ke local people resentful with the self-perception of being
second-class citizens.
Concerning controlling or regulating complex systems, the Law of Unintended
Consequences has a pervasive effect. As a consequence of the irreducibility of com-
plex systems, they cannot always be reliably r egulated or controlled, but since social
arrangements, ma rkets, finance, etc. are man-made, one can strive to r e duce their
complexity to bring them into controllable regions. For example, it may be that s ome
financial procedures and products create sy stems that are inherently volatile a nd un-
predictable exposing society to risk of highly damaging outcomes. In these cases the
contribution of science is to inform the regulators that this particular system, due to
its complexity and unknown to its designers, is inherently unpredictable and undesir-
able. In this case the danger can mediated by simplifying the products and procedures
to make them more predictable and controllable.
In the context of economic systems, neither central planning nor self-regulation
seem to work, but biological regulation (an intricate network of positive and negative
feedbacks, checks and balances on every level) appe ars to be capable of keeping a
complex system in homeostasis, at least for an extended period.
3 Open fundamental questions in Complexity Science
Simple versus comprehensive models.
Complex systems need not be complicated, but in real life they often are. Simple
models are essential to uncover the basic mechanisms and provide insight into fun-
damental questions. However, in order to be able to predict self-fulfilling prophecies,
collective myths, etc.. or forecast the behavior of real systems one often has to go to
more detailed, multi-parameter models. Both approaches have their justification and
they are complementary. However, this kind of “pluralistic” modeling [12] does not
mean the acceptance of differe nt scientific truths, rather it could give more compre-
hensive perspective to the behavior of the system of interest. The models should be
hierarchically related such that previously discovered basic knowledge should serve
as an input into more detailed versions.
When constructing a model of a complex system, the purpo se of the model is
important. In physics we know that any model does not come even close to capturing
all the details of the system. Therefore, we have b e come accustomed to the idea that
“the model should be as simple as possible but not simpler”, but we want the model
to describe some basic features or behavior o f the real system, at leas t rea sonably well.
Thus in our model building we aim for tractability and clarity, by consider ing that
‘models are like maps’ so that they are useful when they contain the details of interest
and ignore others .
So the utility of simple models in describing the complexities
of, for example, poorly unders tood ICT-based social systems is very high. Simple
A passage from Lewis Carroll’s Sylvie and Bruno Concluded illustrates this point: “What
do you consider the largest map that would be really useful?” “About six inches to the mile.”
Will be inserted by the editor 7
models may give deep insig hts in the same way that the simple Ising model pr ovides
useful understanding and quantitative correct predictions on critical phenomena o f
real magnetic systems.
The complexity of a system appears as emergent properties in its often complicated
structure, in how it functions, and in how it responds to exter nal influence of different
kinds. These properties can best be studied empirically from the perspective o f data
analysis. While in natural science well-devised experiments produce the necessa ry
data, in social science the rapidly increasing availability of large scale datasets or
“digital footprints” left by humans in various ICT-r elated systems and services has
created the development of data - or reality-mining (which can also consider e d as
statistical inference). This has become the main source of empirical studies. These
should be accompanied by appropriate modeling. In this respect it is important to
note that the term ‘modeling has different meaning for physicists, statisticians and
social scientists. While in physics modeling is mainly aimed at the understanding
of the underlying mechanisms, in social science (and in statistics) a fitted curve is
already considered to be a model. Here we use the term more in the physics se nse.
Modeling social complex systems constitutes a major challenge because of the
multiple scale s and facets involved. To make progress we need to integrate various
resear ch approa ches of different science disciplines. These include: social psychology
experiments and sur veys, data mining, network ana lysis, complex system and net-
work theory, agent-based modeling and game theory, theory of phase transition and
critical phenomena, intelligent and automated (ICT-based or radio-frequency) data -
collection systems etc. These different approaches when integrated and fused yield
a more comprehensive and complementary picture of the underlying social mecha-
nism and social dynamics at different size-scales from individual, to dyad, to triad,
to group, to community, and to whole s ociety level. Moreover, it is our be lief that the
future of social experimenting lies in the combination of computational and exper-
imental approaches, where computer simulations optimize the experimental setting
and ex periments a re use d to verify, falsify, or improve the underlying ass umptions of
the model.
The aim of simple models is to get better understanding of the so-called “stylized
facts” of the system, i.e., to make simplified, abstracted, or typical observations - in
other words capture so me “essence” of the real system. Of course simple models do not
describe all the details of a system under consideration. Another possible advantage
of simple models is that they may facilitate an analytic trea tment and, thereby, give
better ins ight to the plausible mechanism explaining the behavior of the system of
interest. Simple models can b e extended or made more complicated in a step-by-step
way to capture more details of the system of interest. Mor eover, simple models may
be very use ful in proving that statements made of a system are wrong, i.e., they have
an eminent role in the fa lsification process. On the other hand it may turn out that
the model descr ibes what everybody alr e ady knew, i.e., some common w isdom. In
this case the mo de l, though simple, captures some of the salient features of the real
system. This then serves as a s tarting point for more complicated models, with the
hope of capturing even more features of the real system of interest corr ectly.
Concerning the predictive power of models, it is not necessarily the case that
more complicated models do a better job. In fact it often turns out that simple mod-
“Only six inches!” exclaimed Mein Herr. “We very soon got six yards t o the mile. Then we
tried a hundred yards to the mile. And then came the grandest idea of all! We actually
made a map of the country, on the scale of a mile to the mile!” “Have you used it much?” I
enquired. “It has never been spread out, yet,” said Mein Herr: “The farmers objected: they
said it would cover the whole country, and shut out the sunlight! So now we use the country
itself, as its own map, and I assure you it does nearly as well.”
8 Will be inserted by the editor
els can do a very good job due to their clarity and tractability. Therefore prediction
or forecasting capability is not always a good measure for the usefulness of models,
but rather testable model implications are pluralistic[12]. Furthermore a distinction
between prediction and forecasting should be made: prediction should carry a weaker
but more gene ral meaning, e.g. by predicting types of behavior rather than quantita-
tive fore casting. Also forecasting models, such as weather fo recasting, ar e often based
on known physical laws. In case of social systems one is more inclined to think the
aim of computer simulations is to predict qualitatively the possible behavior of the
Being able to predict the behavior o r forecast the dynamics of a system is fol-
lowed by the possibility o f it being managed. Once one is able to model and predict
the behavior of the system, even qualitatively, it yields understanding of the system
enabling the making of decisions, policies and fur ther development of the system.
This constitutes the ability to manage the system at so me level, which can further be
enhanced by improving the models step-by-step. If on the other hand one is able to
go in mor e quantitative directions and for ecast system outcomes, the model can then
serve as a tool for developing and optimizing the system and its functions.
Micro-macro connection. Choice of variable s.
There are instances in w hich a well established methodology exists to link the
micro and macro descriptions allowing an appropriate choice of variables to descr ibe
a g iven phenomenon. This is not generally the ca se and a methodology is needed to
avoid the temptation of ultrarealistic models in which ir relevant information is in-
cluded: for example, sub-nuclear, atomic or molecular description of water is useless
for wave motion in the sea; and particular car engine characteristics are not relevant
in traffic modeling. Examples of these methodologies include the connection be tween
the atomis tic and hydrodynamic descriptions as succ essfully used in traffic modeling
(the cars being the atoms) or the choice of order parameters based on symmetry
principles as the appropriate variables for the study of continuous phase transitions.
Another example is the methodology of the Renormalization Group in Critical Phe-
nomena that provides a mathematically framework to identify rele vant and irrelevant
parameters by looking for an analytical description of the systems in the spa c e of
the scale transformations. This also leads to a well defined and operational mean-
ingful notion of universality. Likewise, centre manifold theory allows one to identify
the relevant variables and to derive their dynamical equations (amplitude eq uations),
through a multiple time scales analysis, for the description of a system close to an
instability point. Some of these methodologies appear in different contexts with differ-
ent degree of mathematical formaliza tion, but with the same basic contents, such as
the derivation of amplitude equations or the slaving principle of Synergetics (justified
rigorously by normal hyperbolicity theory).
These examples show that success is possible, but finding a framework to solve this
question in general complex systems remains a challenge. The ra tionale b e hind this
question was spelled out by T. Schelling [8] in his Micromotives and Macrobehavior
book:There is a class of important propositions that are true for the aggregate and
not in detail, and that are true independently of individual behavior. Of course this
does not refer to simple statistical properties of a large number of independent units,
but to emerging phenomena that result from their non linear interactions.
The challenge remains to find a methodology or a classification of methods and
protocols for the choice of variables describing complex behavior. Better choices than
intuition or focusing on the variable of interest for the obse rver are needed. Many
financial market models consider that all relevant information is co ntained in prices,
and therefore there is no need to consider anything else. But is this the only relevant
Will be inserted by the editor 9
variable? In models of opinion dynamics the preferred variable of choice is the pro-
portion of p e ople with a given opinion, but it might well be that this is not the most
relevant variable of the dynamical process which should be extracted a posteriori, in
the same way that a directly o bserved qua ntity in a physical instability is not the
dominant amplitude variable for which the dynamics is well characterized.
Beyond the emerge nce of simple collective behavior
Simple collective behavior is an emerge nt property in the behavior of an aggrega-
tion of interacting units that cannot be understood from extrapolation of the prop-
erties of the units. This, for instance is the case of phase transitions in physical
systems. There are well established theories and concepts like broken symmetry [1] to
understand these situations. Flock formation is another example o f simple collective
behavior [13]. But alrea dy in his pioneering paper of 1972 Anderson ide ntifies that
the next stage could be hierarchy or specialization of function, or both [1 ]. Indeed,
there are emergent phenomena that, beyond not being reducible to individual prop-
erties, give rise to hierarchy, multilayered structures and functionalities - a prominent
example being the emergence of organizations and institutions in social systems. We
are still lacking a general theory or a satisfactory and sufficiently general conceptual
framework to de scribe and understand these emergent properties.
Beyond correlations: the se arch for cause-effect relations
Many studies of what are today considered as complex systems have traditionally
relied on blind statistical analysis. The observation of these systems provides correla-
tions of different types. Sometimes these correlations are considered to be some type
of laws of nature ” that should be reproduced by ad-hoc modeling. To go beyond
the knowledge provided by these correlations and to b e able to establish c ause-effect
implications is an urgent challenge. This general question appeared long ago, for in-
stance in the economics liter ature [14]. Rec ent work in this direc tion is in the context
of directed networks inference [15,16]. Still we are far from a satisfactory solution to
this question. On the one hand it requires the identification of mechanisms that are
isolated and implemented in models to investigate their consequences. On the other
hand it also requires new approaches to data gathering and analysis.
Common sense thinking and problem solving often adopts the conc ept of a single
cause and a single effect. It also suggests that small changes in the ca use imply small
changes in the effect. It does no t literally mean that there is a linear relationship
between the cause and the effect, but it means that the system’s behavior will not b e
surprising, and it is predictable, i.e., changes in the parameters or in the structure
of the system do not qualitatively alter its behavior, and the system is structurally
Circular causality in essence is a sequence of cause and effect whereby the expla-
nation of a pattern leads back to the first cause and either confirms or changes tha t
first cause. The concept itself had a poor reputation in legitimate scientific circles,
since it was somehow related to us e vicious circles in reaso ning. It was reintroduced
into science in cybernetics [17] emphasizing feedback. The concept of circula r causal-
ity is reflected also in the theory of reflexivity, an approach promoted in economics
by George Soros [18].
Empirical science is based on the analysis and modeling of data. The explosion-
like development of ICT has resulted in an enormous increase in the data available
for investigation. This is true for traditio nal “har d sciences” but even more so for the
social sciences. Many of our activities leave digital footprints that form huge data sets.
10 Will be inserted by the editor
Our phone and email communication, browsing the internet, using applications like
Facebook, and commercial activities are all documented and can be used for scientific
analysis to provide insight into phenomena and processes at the societal level. This
approach has already made it possible to understand the relationship between the
structure of the society and the intensity of r elationships, the way pandemic diseases
spread and what are the main dynamic laws of human communication behavior. The
availability of data makes it possible to study in detail some of the mo st intriguing as-
pects of complex systems, namely their hierarchical structur e, and how it is related to
the dynamics. Wha t are the laws of “microscopic” socia l interactions? How meso- level
structures form and what is their role? What a re the emergent cooperative phenom-
ena at the societal level? ICT has enabled a new approach called computational social
science and this puts these questions within the scope o f empirical investigations. The
extension of empirical analysis to include massive ICT data, supported by large-scale
multi-agent modeling has provided social science with immediately applicable tools
able to handle issues of major concern. In fact, it gives the hope that mankind may
be able to cope with many pressing issues.
The data deluge related to the ICT brings up s everal challenges. First, much of the
data are not publicly available. Some of them, like mobile phone data are company
property, while others such as much financial data are only available c ommercially.
There have been efforts by sc ie ntists to create an openly accessible po ol of data for
resear ch purposes [19,20], but perhaps the most severe problem of data driven social
science is re lated to this point. In “hard s ciences” reproducibility of results is crucial,
but without open a c cess to data this is not possible in computational social science.
While we should aim at broadest possible availability of data, the production of well
calibrated artificial data sets is one of the important tasks in this field.
Further challenges are r elated to the quality of data. Most often ICT re lated data
are not collected for sc ientific use but, e.g., for billing purpose as in the case of mobile
phone data or for marketing like in the case of point collecting in s upermarkets. In
such cases metadata like age, gender, loc ation e tc. of the people can be assigned to
the data in a rather noisy ma nner. This leads to impure data, with gaps and mistakes.
Cleaning the data and constructing reference data or standards need new techniques
and here massive interaction with social scientists will be necessary.
Of course, the handling of sensitive data raises ethical proble ms [21]. At the mo-
ment there is only limited regulation in this respect. Two opposite opinions have been
formulated: (i) no extra regulation is needed in addition to the general legal frame-
work; and (ii) there is need for institutiona l solutions similar to those in genetics and
traditional social science. A thorough study by the National Academy of Sciences,
US [22] supports the latter view.
Even in everyday life the data deluge has changed our attitude to information.
While previously searching for information was typical, today selection has become
most important. This means in science that data mining and processing techniques
have become crucial for the development of the field o f complexity scie nc e. While
the question of dimensiona l re duction of the data to arrive at useful information
remains a central problem (as is usual in empirical science), the next foreseeable fron-
tier is a complementary approach that implies a shift from data-driven modeling to
question-driven data- gathering i.e., the goal is producing or gathering data to ans wer
a specific question. This is in the spirit of c lassical experiments designed to obtain
data to test theoretical predictions. True validation o f models, as opposed to models
fitting raw data, requires comparison of model implications (quantitative or qualita-
tive) with data obtained under the conditions and assumptions of the model. Such
experiments are naturally designed in so me virtual environments, like internet g ames
or in electronic social networks, and they are part of the new undertaking in so c ial
experimenting [23]. Of course, ethical issues have to be handled with appropriate
Will be inserted by the editor 11
care. We refer to Section 7.2 for a more detailed discussion of the platforms for social
computing and web-gaming.
Ensemble modeling and Data assimilation
There are fields such as weather prediction or climate re search in which the fun-
damental microscopic laws are well k nown and established but given the huge range
of r e levant scales , meso and macro models are used that implement in different ways
large-scale effective interactions and parametrizations of the same basic phenomena.
In these fields a common practice involves probabilistic forecasts based on the com-
bined results of different models or different specification of a g iven model. This
methodology is used in a variety o f fields [24]. An important methodological question
is to which extent this pluralis tic modeling [12] or c ombination of forecasts [25] is
conceptually justified beyond purely statistical considerations, in other fields, such
as social phenomena modeling. Pluralistic mode ling should, however, not mean that
different models origina ting form different basic concepts can be simultaneously co n-
sidered. A related methodological q uestion is the possible general use of the data
assimilation procedures of atmospheric modeling [27]. The basic idea is combining
forecasting and observation for initial conditions in dynamical modeling: Forecasting
at a given time has to be combined with large scale observations at that time. Another
methodology recently developed in climate (see e.g, [26]) is bas e d on analyzing the
hidden information in the dynamics of weather simila rity (network links measured by
cross- c orrelations) between pairs of locations. This yields an evolving network char-
acterization of the climate that was found useful for example in better unders tanding
of the El-Nino pheno mena.
From Data to Dynamical Models
Figure 1 illustrates the scientific perspective of complex systems methodology.
It begins with data from which scientists recons truct phenomenological models. For
example, Kepler constructed a phenomenological model in which the planets sweep
out eq ual areas in e qual times which Newton formulated as a theory of planetary
motion able to reproduce this phenomenology. In the case of the motion of two bodies,
Newton’s Laws produce equations that can be so lved explicitly making it possible,
for exa mple, to predict precisely where a cannon ball will land. In the three-body
case the equations cannot be integrated and the system is chaotic. Nonetheless the
spatio-temporal behaviour of the sys tem can be simulated by iterated computation
providing an augmented phenomenology (Figure 1, bottom right). The objective in
this modeling is to pr oduce an augmented phenomenology whose statistical difference
from observation, , is as small as possible (theoretically zero for a perfect model). In
most cases simulations can at best sample the space of all system trajectories around
given initial conditions with an error, , which measures the difference between the
statistical distributions of the s imulated trajectory and the statistical distributio ns
of the data. Since each iterated calculation in a simulation of a system sensitive to
initial conditions creates error, increases with time.
As a social systems example consider the people evacuating a building in an
emergency. Helbing [80] observed the motion of people in crowds and created a phe-
nomenological model of the ways people move with res pect to each other. Using
this phenomenological model Helbing used agent-based co mputer simulation to cre-
ate an augmented phenomenology for this system (Figur e 1, bottom-centre). Helbing
went on to formulate theoretical models of pedestrian flows (centre top) permitting
spatio-temporal simulations (mid right) to c reate another augmented phenomenolog y
(Figure 1, bottom-right). In this case both the ag ent simulation and the theoretical
model gave an augmented phenomenology with small error. In fact Helbing went on
to use this new science to assist the authorities to redesign Mecca for the Hajj pilgrim-
12 Will be inserted by the editor
measures the statistical distance
between distributions of trajectories
of the augmented phenomenology
and distributions of observations.
Theoretical Model
Fig. 1. The complex systems methodology for reconstructing models and theory from data
age which was subject to fatal accidents with large numbers of people being trampled
as the dynamics of the crowd changed. The redesign was very successful and many
lives have been saved [81]. This is one of the major success stories of complex systems
4 Interconnected multiple scales networks
In 1998 Watts and Strogatz [28] pres ented the first evidence of what they called
complex networks. Watts and Strogatz realized that patterns of connections in the
neural network of a worm, in the power-grid of the United States, and in the network
of co-appearances of actors in movies, showed common features that could not be
explained by simple mathematical models (the two extremes of total or de r or total
randomness being the most simple). The presence of short- cuts made these networks
extremely small while at the same time kept track of their ordered origin through
a larger than expec ted clustering coefficient (related to the existence of triangles in
the network). On the other hand, as shown slightly later by Barabasi and Albert
[29], the distribution in the number of c onnections coming out from given node (e.g.
computers in the Internet, pages in the world-wide-web, or number of papers authored
by scientists) show skewed distributions, which means that there e xist some actors
in the ne twork s that are highly connected, which were c alled hubs. These two works
were the seed of a new field of research.
Nowadays, it is understood that most complex systems show emergent dynamical
properties which are inherently rela ted to the topo logy of the underlying network
of connections among the constituent parts of the system. During these fourteen
years we have witnessed how these ideas have been applied to a myriad of problems
ranging from the cell scale of biochemical networks to the scale of world population
communicating through the information and communication technologies [30,31].
Up to now we have considered mostly that a given network description is good
for a given problem. If we plan to understand how proteins interact then we lo ok
at the protein-protein interaction networks, and if we want to prevent damages in
the energy distribution we a nalyze the fea tures of the power-grid networks across the
Will be inserted by the editor 13
But the question that arises now is to which extent these networks are really
independent? Is the distribution of electr ic ity across the p ower-grid independent of
the transportation of other energy re sources such as gas [32]? Is it independent of
the tra de r e lation between countries a s illustrated by the recent diplomatic iss ue s
between former soviet republics? Is it independent of the communication network
that connects power stations and distribute energy according to generatio n and load?
The answer is clearly not - these networks are not independent. The communication
network among power stations depends on the stability of the power-grid and the same
for the other examples [33]. Buldyrev et al [33] recently introduce d a mathematical
framework, based on percolation theory, to study the robustness of a network formed
by interdependent ne tworks.
But this interrelation not only affects technological or transportation networks,
it also affects social ne tworks. To how many social networks do we belong? From
the traditio nal classification of so cial networks we can identify friendship networks,
kinship networks, profes sional networks, and in a more modern framework we belong
to different types of online ne tworks. Networks can relate customers among them-
selves: people who purchased the same items, people that like and comment on each
other’s pictures, and people coopera ting in online games together. How are all these
networks intertwined and how do they affect each o ther? Online social networking is
changing our way of living, sharing, or e ven feeling. Perhaps one of the main draw-
backs of Facebook is that the definition of a “friend” just means someone you are
connected to. Whenever you, as Facebook user, receive a message saying: ... wants
to be your friend”, it can be your mother, your student, your thesis advisor, an ex-girl
(or boy) friend, o r your online war allies. On the contrary, a ne w proposal by Google
(Google+) trie s to change this featur e with its “circles”, in which one can choose the
type of relationship with the other users and hence separate the different networks we
belong to. Those are clearly two different perspectives on something that for people is
simple “who is connected to who”, but for scientific reasons it is extremely important
to distinguish these types of connections that make up our social world.
Even s ocial and technological networks are strongly interconnected, e.g. how much
our connected social world depends on electricity resources. Among the most studied
networks over the years we find the Internet and the World- Wide-Web. The first is
identified with the set of hard wired connections (and now includes wireless pr otocols
as well) between Internet providers and users that forms the basic infrastructure for
all current communication technologies. The second corresponds to the set of pages
which contains the information. But nowadays, the flow of information is not only
provided in one direction from the web servers to the final users, but streng thens
relations among use rs or among communities at different scales through different
ways of communication: emails, chats, messaging, blogging, sharing, and so on. But,
closing the circle, new social interactions are still on top of technological networks
that are vulnerable and can be monitored and even censored. And one of the most
clear examples took place during summer 2011, since mes sages fueling the London
riots were mainly broadcast from smartphones using a private network which encrypts
Additionally, these networks form aggregates of users that can be analyzed at
different scales, because the questions one poses and hence the answers one gets are
quite different. Studies have bee n performed on online users that are in the same
class-r oom or at college level [34]. O ther studies focus on the spreading of political
ideas through Facebook or Twitter (in the Arabian co untries, in Finland or later in
Spain and the 15M movement of “the indignados” [35] ( Finally,
other empirical analys es have focused on the widespread use of messaging services as
Microsoft Messenger [36].
14 Will be inserted by the editor
This importance of different scales is also observed in the propagation of dis-
eases. Spreaders (humans) can travel long distances following different transportation
modes: long and fast jumps made by flights through the airport networks; medium
range trips are mostly by car or bus then related mainly to road networks ; while short
and frequent trips corres pond mainly to commutation networks of public transporta-
tion. In this case transportation networks at different scales interact. This is how
real infections spread, but a curious example about the relation between networks
of different origins is found in the recor ds from the web searches during the H1N1
virus propag ation. These searches, ma de by people with some of the symptoms are
geolocalized and combined give hints on how the disease is being propagated.
In the near future researchers on complex socia l networks will focus on networks
at different scales, with intertwined different meanings. Pro bably the picture will not
be that simple. Nowadays we know that networks are dir ected, weighted, adaptive,
space embedded, interdependent and so on, and furthermore they are also dynamic.
This includes the dynamics taking place over the links [31,37], and also the inherent
dynamical nature of the connections themselves, with coevolution processes of the
dynamics on the network and the dynamics of the network [38,39]. This will introduce
further ingredients that the new science of complex networks will face in the future
For all these reasons, complex networks theory has to develop new tools, new
measures, and new models that account for all these new ingredients, na mely the
interrelations between networks of different origins, networks interacting at different
scales, cross correlation be tween dynamics o n networks and the dynamics of ne twork
topology, including dependence on patterns of connectivity. Collaboration with social
scientists and ICT-researchers will be cruc ial in de veloping a new framework for the
final understanding of new social features and behaviors, and to construct new so-
cially inspired communication resilient technologies. Also, fo r more details on current
challenges on complex networks resear ch see the paper on ne twork s by Havlin et al
in this volume [41].
5 Information aggregation and processing. Social learning
At different social levels, from the family, to international coalitions, to the global
human society, we need to take collective decisions that shape our future. Taking
a good decision ultimately depends on our ability to aggregate information that is
widely dispersed. This implies making choices on what information we pay attention
to and how we discriminate between what we consider r elevant or accurate infor-
mation and what we consider background noise. These processes depend on very
different issues, some of them strongly technological such as information propagation
and information ava ilability, and others strongly social such as trust. Needless to say,
the problem of informa tion a ggregation takes a completely new perspective in the
light of the societal changes associated with the new Informa tion and Communica-
tion Technologies, which imply different, faster and broader ways of communication
as described in the next sectio n. Today the flow of information occurs at multiple
temporal and spatial scales. The compelling task has changed from accessing infor-
mation to selecting infor mation avoiding misinformation and disinformation. How do
we do this sele c tion and aggregation, and what are the consequences?
From the above perspective, social learning can be defined as the ability of a
population to aggregate information [4 2]. This process drives phenomena like opinion
formation or political changes (either smooth or deep changes). Individual learning
follows, in a traditional setting, from global-local competition: the competition be-
tween global information received through media, advertisements, etc. and informa -
Will be inserted by the editor 15
tion learnt and adapted from one’s social circle. This competition has today different
characteristics due to the new extent and meaning of the social circle: the ease of
interacting with any other individual redefines the concept of social circle. An impor-
tant change of attitude also exists with respect to news information: many people now
search actively for news in different ways on the internet instead of passively watching
TV news or habitually reading the same newspaper. The challenge is to understand
that processes of information aggregation in a society are not necessarily driven by
information hubs that process information and broadcast it, but are strongly coupled
in a society that aggregates information in different collective processes driven by
pair-wise, group and multi-institutional interactions. A clear example in this context
is the change from the Encyclopedia Brita nnica to the Wikipedia.
A basic question is the possible shift from a society that ex ploits social learning
by aggregated information and feedback processes to improve this le arning, rather
than relying on the traditional specialists or experts. An important challenge in this
context is how to avoid or how to identify information cascades or rumor-spreading
that amplify errors or misconceptions to a globa lly accepted false truth. There is the
the wisdom of crowds, but crowds are not always wise. Also, there is the que stion of
different processes for aggregation of information about facts and about interpretation
of the facts, that is the aggr egation of meaning. Collectively these questions are behind
two of ten top social-science questions listed in [57]: How can we aggregate information
possessed by individuals to make the best decisions?, and How can humanity increase
its collective wisdom?
Information aggregation and processing is not specific to human societies and is
a common process in many other natural complex systems, as for example in bird
flocking or fish s chooling. In this context a recent experiment [5 8] ha s demonstrated
genuine wisdom of the crowd [59] showing that larger s hoals of fish make more ac-
curate and faster decisions (avoiding a predator) than smaller shoals. The design of
the e xperiment is also an example of data gathering to test modeling predictions or
to answer a g iven question, beyond the use of raw available data. What we learn
about information processing by decentralized information communication in natu-
ral systems is also a guide to designing new and alternative information processing
systems, searching for the emergence of intelligent behavior from simple interaction
rules. Research in Swarm Intelligence is an example in the dire c tion of implementing
self-organized coordination of many individuals by decentralized information commu-
Finally, knowledge aggregated from complex self-organizing human or natural sy s-
tems opens up the challenge of the implementation of unconventional computatio nal
principles base d on complex dynamical systems [60], such as reservoir computing
[61,62] or information processing base d on complex systems dynamics [63].
6 Socio-Technical Systems
The term “socio-technical systems” refers to the interaction between technologies and
human social behavior. Psychologists initially re c ognized this interaction in the early
50’s. In a pioneering work E.L. Trist and K.W. Bamforth [64] studied the social con-
sequences of the adoption of a new production technology in coal mining leading a
productivity fall. In this general context one can a sk what happens to a society whe n
new forms of communication appear. This question, which is fundamentally impor-
tant and has far-reaching implications, is what the Information and Communication
Technologies (ICT) has brought about over the last decade. ICT has radically and
unforeseeably changed society as a whole. This is true not only in highly industrial-
ized countries as shown for example by the large impact and penetration of mobile
16 Will be inserted by the editor
phone networks in developing countries in Africa. At first sight, these changes can b e
attributed to the actions of individuals and the availability of new channels of commu-
nication that transform basic social processes: (i) face-to-face encounters have become
less critical than in the past, (ii) the dynamics o f building and strengthening relation-
ships have evolved by taking advantage of ICT, and (iii) new ICT-mediated groups
and communities have emerged, by overcoming typical limitations such as distance
or lack of a common platform. In addition, entirely new ways of collective human be-
haviour have appeared, such as those collaborative a nd sometimes co nflicting actions
exemplified by Wikipedia.
However, this description is critically inc omplete because it fails to recognize that
individuals, society, and ICT are deeply intertwined in a dynamic feedback process,
where individuals adopt new communication channels to form and join groups that
change in identity and size, thereby restructuring the whole of society. Simultaneously,
ICT providers develop new channels of communication, some of which fail while others
become enormously popular. Indeed, unpredictability is a characteristic feature of
these developments. Popular channels such as WWW and SMS were not originally
designed for the purposes they serve today. Entirely new platforms for ICT-media ted
social interactions, for example Facebook, have emer ged “out of the blue”. They have
gained mass popularity in a very short time and transformed the social behaviour
of individuals in a number of unexpected ways. An example is the role of Twitter in
mass movements such as Arab Awakening of 2011 or the Spanish 15M movement [35].
In our view, the fundamental challenge for future social ICT is to overcome the acute
lack of unders tanding of the driving forces and mechanisms of this complex system
of interactions b etween individuals, society, and ICT.
This deficiency requires developing systematic means of exploring, understanding,
modelling and possibly even controlling systems where ICT is entangled with social
structures. In particular, there is need to focus on the behavioural patterns, dynam-
ics and dr iving mechanisms of social structures whose interactions are ICT-mediated,
from the level of individuals, dyads, and triads to the level of groups, communities,
and large-scale social systems [65]. The research approach necessa rily has to be based
on c ombined exp ertise in complex systems, computational analysis and modelling,
and social sciences. In contrast with studies that start from extremely s implified as-
sumptions concerning social dynamics and conc e ntrate on finding structural features
of social systems, it should be emphasize d tha t ICT networks are dynamic sy stems o f
interacting humans and groups, and should thus fully utilize the theories and methods
of the social sciences.
New ICT also puts public goods problems in a new perspective. An example is
water management or waste management. How do we take decisions on our individual
behavior in these issues in a globally interdependent society when we receive full
information on daily situation and global consequences? Likewise, smart grids are
now designed so that centralized decisions for better energy management are taken in
the context of data collected from a large number of distributed sens ors. The challenge
is to introduce into the design the adaptive behavior of the users and to explore self-
organized grids in which properly ag gregated data from the sensor s is made available
online also to the users [43].
7 Data gathering, Participatory Sensing and Social Computing
7.1 Citizen Science
The issue of sustainability is now at the top of the political and societal agenda and is
considered to be of extreme importance and urgency [44,45]. There is now overwhelm-
ing evidence that the current organisation of our economies and societies is seriously
Will be inserted by the editor 17
damaging biological ecosystems and human living conditions in the very short term,
with potentially catastrophic effects in the long term. In a recent statement from the
head of the European Environmental Agency, there is a r e alisation that only through
bottom-up actions we can deal with today’s challenges: “The key to pro tec ting and
enhancing our environment is in the hands of the many, not the few.... That means
empowering citizens to engage actively in improving their own environment, using
new observation techniques...” [46].
The enforcement of novel policies may be trigge red by a grassro ot approach, with
a key contribution from information and communication technologies (ICT). Nowa-
days low-cost s ensing technologies allow the citizens to directly assess the state of the
environment; social networking tools allow effective data and opinion collection and
real-time information spreading processes. In addition, theoretical and modeling to ols
developed by physicists, co mputer scientists and sociologists have reached the matu-
rity to analyse, interpret and visualize complex data sets. A techno-social system,
acts like a lens that captures information from the environment: one has to explore
the pe culiarities of having human agents as sensing nodes, the role of noise sources
at different s cales, the effect of opinion bias, information spreading in the community
supporting the techno-social system, network effects, and so forth.
Devices employed in the connection to communication networks have converged
in siz e and technological standards. Cell phones have integrated many functions
traditionally accomplished by personal computers. This progress while being use-
ful, yields also new kind of risks and challenges such as epidemics of viruses and
malfunctions[47]. In turn, computer manufacturers have privileged products de signed
for an easy mobile usage, such as new generation tablets. Moreover, cell phones and
PCs incorporate sensors of increasing accuracy: GPS sensors, cameras, microphones,
accelerometers, thermometers are already standard equipment in many devices. Net-
works have also accompanied this process, by expanding the availability of an Internet
connection throughout daily life. Op en-hardware platforms, such as the well-known
programmable microcontroller based Arduino, will also facilitate the task of tak ing
an input signal from the e nvironment, process it, and deliver it thro ugh the I nternet
at a low cost.
The large number of sensors deployed is alre ady turning urban areas into “ smart
cities”, that is, intelligent and complex organisms able to process the sensors signals,
visualise them and possibly trigger the automatic e xecution of appropr iate actions
(see Michael Batty et al. contribution in this volume). The mobile, powerful, and per-
manently connected equipment described above makes any citizen a potential source
of sensor data about her/his environment, with little or no scientific skill required.
Participatory sensing experiments involve communities of such individuals in the mon-
itoring of a particular issue, e.g. the quality of a metropolitan environment [49] or
the redevelopment o f urban areas. This is not entirely new, since numerous “citizen
science” initiatives have be en already launched in areas ranging from ornitho logy to
astronomy, with or without the help of sensors. A recent trend is represented by the
integration of crowdmapping and participatory sensing through the web and several
important initiatives have been car ried out, e.g. to monitor the spreading of the In-
fluenza A virus
or social mobilization[48]. It is important to remark how this data
gathering activity is very relevant for the so-called data-driven simulations, i.e. sim-
ulations of complex systems whose predictability accuracy crucially depends on the
interplay between the goodness of the modeling scheme and the possibility to moni-
tor several observables to recalibrate in real-time the evolution of the system under
investigation. In addition, online platform, such as, have shown in
18 Will be inserted by the editor
practice how the data collection activity and its vis ual representation reinforce them-
selves. The access to both personal and community data, collected by users, processed
with suitable a nalysis tools, and re-presented in an appropriate forma t by usable com-
munication interfaces, has the potential of triggering a bottom-up impr ovement of
collective social strategies as well as stimulating fundamental shifts in public opinion
with subsequent changes in individual behaviour and pressure on policy makers[50].
Particular events, such as the nuclear c risis following the 2011 earthquake in Japan,
have demonstrated that involving citizens in the environmental monitoring activity
is an e ffective method to build accurate risk maps . The participatio n of users in the
monitoring aects both the resolution and the quality of the data collected. While
traditional sensing generally involves a s mall numb er of highly controlled observation
points, distributed sensing relies on the possibility of gathering large amounts of da ta
from ma ny uncontrolled sources, which cannot ensure high data quality standards;
however, by means of statistical methods together with the possibility of storing and
post-proce ssing large datasets, this quality gap with respe c t to traditional sensing
can be overco me. The refore, the analysis tools should be able to detect and filter out
deviations due to sensors misuse or to biases introduced by the users themselves.
7.2 Platforms for ICT-based experiments
In the last few years the Web has been acquiring the status of a platform for social
computing, able to coordinate and exploit the c ognitive abilities of the users for a
given task. One striking example is given by a series of web games [51], where pairs
of players are required to coordinate the assignment of shared labels to pictures [52].
As a side effect these games provide a categorization of the images content, an ex-
traordinary diffcult task for artificia l vis ion systems. More g enerally, the idea that
the individual, selfish activity of users on the web can possess very useful side ef-
fects, is far more general than the example cited. T he techniques to profit from such
an unpre c edented opportunity are, however, far from trivial. Specific technical and
theoretical tools need to b e developed in o rder to take advantage of such a huge quan-
tity of data and to extract from this noisy source solid and usable information. Such
tools should explicitly consider how users interact on the web, how they manage the
continuous flow of data they receive, and, ultimately, what are the basic mechanisms
involved in their brain activity. In this sense, it is likely that the new ICT- mediated so-
cial platforms, could ra pidly become a very interesting laboratory for social sciences.
In particular we expect the web to have a s trong impact on the s tudies of opinion
formation, political and cultural trends, globalization patterns, consumers behavior,
marketing strateg ies. A very original example is represented by Amazon’s Mechanical
Turk (MT) (, a crowdsourcing web service
that coordinates the supply and the demand of tasks that require human intelligence
to complete. It is an online labor market in which users perform tasks, also known
as Human Intelligence Tasks, proposed by ” e mployers” and are paid for this. Salaries
range from cents for very simple tasks to a dollar or more for more complex ones.
Examples of tas ks range from categorization of images, the transcription of audio
recordings to test websites or games. MT is perhaps one of the cleare st examples of
the so called crowdsourcing and thousands of projects, each fragmented into small
units of Work, are perfo rmed every day by thousands o f users. MT has o pened the
door for exploration of processes that outsource computation to humans. These hu-
man computation processes hold tremendous potential to solve a variety of pr oblems
in novel and interesting ways. Thanks to the possibility of recruiting thousands of sub-
jects in a short time, MT represents a potentially revolutionary source for conducting
experiments in s ocial science [53,54,55]. It could become a tool for rapid development
Will be inserted by the editor 19
of pilot studies for the experimental application of new ide as. As a starting po int for
this new idea of experiments, the blog
already presents a review of the results of a series of classic game theo retical exper-
iments carried out on MT [56]. Despite its versatility MT has not been conceived
as a platform for expe riments. This is the reason why it is important to develop a
versatile platform to implement social games. Here the word game is intended as
an interaction protocol among a few players implementing a specific ta sk and it is
used as a sy nonym of experiment. The development of such a platform has to sa t-
isfy a certain number of requirements among which high modularity and flexibility,
synchronous (i.e., r e al time) and asynchronous interaction modes, robusteness with
respect to heavy loads to process and store a continuous data flow. The adva ntage
of this kind of experiments is that every useful piece of information and detail of the
evolution will be fully available and leveraged for benchmarking as well as for the
modelling activity. Moreover the effects of social interactions can be observed with
a larg er statistical basis and in a more controlled environment. It should be stressed
that these ICT-based experiments are truly general purpose since through them one
can investigate complex phenomena in a wide range of disciplines including (but not
limited to) social sciences, economics, psychology and linguistics. In the framework
of European project EveryAware
a first prototype of such a platform is being real-
ized, dubbed Experimental Tribe ( (ET). ET is intended as a general
purpose pla tform that allows the realization of a very large set of possible g ames. It
has a modular structure through which most of the complexity of running an exper-
iment is hidden in a complex Main Server and the experimentalist is left with the
only duty of dev ising the experiment as well as a suitable interface for it. In this way
most of the coding diffculties related to the realization of a dynamic web applications
are already taken care by the ET Ser ver and the realizatio n of an experiment should
be as easy as construc ting a webpage with o ne of the many online services for it.
The benefit is twofold: on the one hand, it allows virtually any researcher to realize
his own experiment with minimal effort, paving the way of the use of the web as a
standard “laborato ry” to perform experiments. O n the other hand, it can be a strong
“basin of attraction” for people willing to participate to experiments, making in this
way recruitment much more easier than for single- experiment platforms.
8 Systemic risk, extreme events and predictability
Extreme events both in nature and society, such as ea rthquakes, landslides, wildfires,
stock market crashes, the de struction of very tall tower buildings, engineering failures,
outbreaks of epidemics etc. may appea r to be surprising phenomena whose occurrence
does not follow any rules. Of course, such kinds of extreme events are rare, but they
influence our everyday lives dramatically. Can we understand, assess, predict and
control these events?
Complex systems theory offers a new perspective to understand the mechanism of
the emerging patterns. As a consequence of natural and social crises, the occurrence
of rare large extre me events are now the focus of extensive mathematical analysis
8.1 Widening the Limits to Prediction of Extreme Events
It is common to hear questio ns such as what is the probability of having a big earth-
quake in Iceland w ithin a year?” or “how large might a possible stock market crash be
20 Will be inserted by the editor
tomorrow?”. The study of earthquake eruptions, the onset of epileptic seizures, and
stock market cra shes traditionally are investigated by very different disciplines which
differ very much in their scientific culture. The co mplex sy stem approach emphasizes
the similarities and offers some common methods to predict the behavior of these
systems, and/or understand the inherent limits of their predictability [68].
Standard sta tis tical procedures neglect data p oints deviating greatly from oth-
ers, the so -called called outliers. Extreme value analysis uses statis tical methods to
analyze rarely occurring events. Typically, ex treme events occur in the tails of proba-
bility distributions as a function of the “size” o f the events (such as energy, duration
etc). Emil Gumbel (1891-1966) a famous pacifist, contributed significantly to the es-
tablishment of statistical methods to describe extreme deviations from an average”
behavior. As he wrote: “It seems that the rivers know the theory. It only remains to
convince the engineers of the validity of this analysis.”
Extreme value analysis, a branch of mathematical statistics , estimates the proba-
bility of extreme floods, large insurance losses, market risk, freak waves, tsunamis, etc.
While the Gumbe l distribution shows a light-tail (expo ne ntial decay), other classes
of “extreme value distributions” behave differently. Distributions of ea rthquakes and
avalanches have extreme value statistics described by power-law tails . These imply
that extreme events occur much more frequently than expected. For example, the
crash of the stock market on Bla ck Monday was a 35σ event, whe re σ is the standard
deviation of the Dow Jones Index on a logarithmic scale. Knowledge of the size distri-
bution of floods, storms, earthquakes is highly important for the insurance bus iness
and for the risk assessment of financial derivatives.
In the context of power laws, Sornette has suggested the possibility of “tra nsient
organization into extreme events that are statistically and mechanistically different
to from the rest of their s maller siblings” [79]. He calls these dragon kings where
“Often, dragon kings are associated with the occurrence of a phase transitions , bi-
furcation, catastro phe, tipping point, whose emergence organization produced useful
The theory of complex systems s uggests that extreme events may b e predicted by
detecting their precursors, and that there are methodological similarities for analyzing
and modeling different “critical events” occurring in physical, biological and social
phenomena. There are initial promising results and many open problems.
8.2 Dynamical models of extreme events
To be able to control and manage extreme events we should understand the generating
mechanisms (and gener ative models) of the phenomena. One possibility is to say
that big earthquakes are nothing else but small ea rthquakes that do not stop. The
consequence is that these critical events would inherently be unpredictable, since they
don’t have any precursors. This approach is called self-organized criticality (SOC) and
was championed by Per Bak [69]. Self-orga niz e d criticality suggests that the same
effect may lead to small, but also to very large avalanches, so the outcome is not
really predictable. A famous toy model is the sand-pile model [70].
According to Sornette’s arguments [66] catastrophic events, or at least a class of
them, result from accumulating amplifying cascades. Based on the hypothesis of this
theory of intermittent criticality, many stock market cra shes are generated by a slow
building up of “subterranean fo rces”, and their precursors may be detected. Were this
hypothesis true, the predictability of these events may be possible.
Uncompensated positive feedback can be a mechanism for crashes. Positive feed-
back seems to be a general mechanism [71] behind the eruption of earthquakes, stock
Will be inserted by the editor 21
market crashes, hyperinflation, and epileptic seizures. The lack of the stabilizing ef-
fects of negative feedback mechanisms may lead to catastrophic consequences. If there
are no mechanisms to compensate for the effects of higher-than-linear positive feed-
back, the processes lead to finite-time singularities.
In an ec onomy there ar e many feedbacks that drives it towards equilibrium be-
tween demand and supply. There are many positive feedbacks, such as those due to
our susceptibility to imitate each other’s behaviour, and these can lead to explosive
growth in price s, followed by the inevitable bursting of the bubble. Equilibrium theory
works well when negative feedback effects have stabilizing effects to positive feedback
changes. While in normal situations the activities of buyers and sellers neutralize
each other , in critical situations there is a cooperative effect due to the imitative
behavior of everybody wanting to buy since everybody else has already bought, and
the positive feedback is higher-than-linear. Such super-exp onential increases, due to
irrational expec tations, canno t continue for ever and the increase is unsustainable.
Consequently, it should be followed by a compensatory process, i.e., a stock market
9 Control and management of complex global systems
A pervasive challenge for complex systems science is to control or manage complex
systems. Here is a list of topical examples:
Financial System Social Unrest Economy
Health Service Famine Relief Epidemics
Electricity Pricing Schemes Demographics Climate
The words control” and “management” may be used interchangeably, but many
social scientists dislike the word “co ntrol” which carries overtones of authoritarianism
and prefer the word “management” which c onveys a mor e benevolent approach.
There is an important distinction to make, however, between two forms of control.
In the strong form of control, the objective is to make the trajectory of the system
follow some desired track or reach some targe t [72]. In the weak form of control, the
objective is to make the probability distribution for the trajectories of the system
follow s ome des ired track or reach some target (in stochastic control, some integral
with respect to the proba bility distribution is usually optimised).
This distinction is crucial. As soon as the dynamics of a deterministic system shows
sensitive dependence on initial conditions (“chaos”), control of a trajectory is likely
to re quire feedback with higher gain than the maximal Lyapunov exponent (though
examples can be made where arbitrarily small carefully chosen gain matrices suffice),
which may involve unrealistically high observational power and actuator response.
Similarly, control of a stochastic jump system requires control response time to b e
shorter than typical waiting times. In the case of diffusive systems, a simila r criter ion
is required but depends on the a ccuracy with which one wishes to track.
In contrast, chaos or stochasticity a re good for making the probability distribution
for the trajectories of a system relax rapidly to one that is unique. Thus this proba-
bility distribution may be controlled by much slower observation/actuator feedback
than the individual trajectories. Also much less detailed o bservations and controls
may suffice, thus avoiding Orwell’s “big bro ther nightmare a nd making them more
acceptable to our “free” society.
This section concentrates on the control of probability distributions for trajecto-
ries. They have been christened “space-time phases” [73]. I n the next few paragraphs
22 Will be inserted by the editor
we propose a r ole for substantial new mathematical developments. For some back-
ground, see [74,75].
The simplest context in which to begin is probabilistic cellular automata. These
consist of a network of units whose states update in parallel in discrete time according
to probability distributions that depend on the current state of the whole network but,
conditional on the current state, the distributions for different units are independent.
In contrast to much of the literature (e.g. [76]), there is no ass umption here that the
network is a r e gular lattice, that the units are identical, or that the dynamics are
Under suitable conditions, the operator representing the e volution of probability
distributions for the state of the network is an eventual contraction in a suitable
metric, and this leads to exponential convergence to a unique probability distribution
for the trajectories. The resulting space-time phase depends smoothly on parameters
of the model, thus its dependence on feedback control laws can be studied. Given
design objectives, one could then seek feasible control laws to bring the statistical
behaviour of the trajectories close to the objectives.
Although the above holds for all indecomposable systems with finitely many units,
a more appropriate approach for la rge systems with some strong interdependence of
their units is to consider them as part of an infinite system (just as in equilibrium
statistical mechanics). Then the possibility of non-unique space-time phas e emerges.
As parameters are varied the system may jump from o ne space-time phase to an-
other that is far away. This is a reflection of the popular notion of “tipping point”.
Even without parameter va riation, the orig inal finite system may best be described
as making random transitio ns between two or more such phases. The ways the set o f
phases can depend on parameter s is a fairly wide open question: some semi-continuity
results hold, but there is a great need for an analogue of the bifurcation theor y for
simple attractors of deterministic dynamical systems, so that we could understand
what are the typical qualitative changes in the set of phases. Going further, could
controls be de signed to collapse the set of phases into a desired unique one? This con-
nects with another branch of deterministic dynamical systems theory called “ergodic
optimisation” in which the aim is to stabilise an invariant pro bability distribution on
an attractor differing from that naturally chosen.
Once the theory for probabilistic cellular automata is well developed, it will be
natural to seek to e xtend it to more realistic classes of system, for example continuous-
time stochastic jump processes, systems of mobile units where the strength of interac-
tion depends on distance in physical space, and deterministic systems with sufficiently
chaotic dynamics.
Let us turn now to the examples.
Redesign of financ ial regula tion is urgent. The current space-time pha se has bub-
bles and crashes a ll the time, e ven if the recent banking crisis has been the worst ever
and national debt crises may overtake that. We need to move the financial s ystem to
space-time phases in which bubbles are deflated before they grow too big and debt
is not allowed to grow to unserviceable levels. We need to analyse the effects of pro-
posed policies like separating re tail and investment banking, introducing a tax o n all
financial trans actions, imposing time de lays on trades. The models need to include
the decision-making behaviour of real people and their confidence, not just money.
Social order can be very fragile as seen in the recent riots in England. Almost
certainly this is a system with (at least) two phases: order and a narchy. We must
understand which management strategies make social order more stable. The models
need to include such factors as feeling of belonging and feeling one has a future.
Stability is not the only des irable feature, of course; one must also address the nature
of the res ulting social order. Social order is a subject with a long history, e.g. [77], and
Will be inserted by the editor 23
forms the core of sociology, yet we believe that the time is ripe for serious advances
of a mathematical nature.
Economies are notoriously difficult to manage. The business cycle and its more
extreme versions like recession correspond to long-range corre lations in space a nd
time of the space-time phase. It might be that this is a natural result of seeking to
maximise productivity, just as seeking to pass heat fas ter through a fluid layer leads
it to form convection rolls. Is there some control strategy which can achieve as much
productivity without the large scale o scillations? Is there a c ontrol strategy that can
achieve it with close to full employment, a goal that would be fulfilling for most
people and surely could be more productive? One of the issues this example raises
is that planners often require more than one, often incompatible, objective. Thus for
example, the USA Federal Reserve sets the nation’s monetary policy to promote the
objectives of maximum employment, stable prices, and moderate long-term interest
rates”, but admits that “tensions among the goa ls can arise in the short run”. It may
be that these goals are incompatible in not just the short r un but for e ver and, in the
terms of Herbert Simon, have to be satisficed [78].
Several countries are having immense trouble with the organisation of their health
services. Yet simple measures such as hygiene, prevention and control could substan-
tially shift the space-time phase to one that is much better.
We are struck by the images of famine in Somalia, but drought and malnutrition
are a regular feature of the probability distribution there and indeed in other parts
of the world too. Is there a way to manage food production and dis tribution that
would avoid the extreme of hunger and the opposite extremes of obesity in the USA
and UK? The problem is linked to demographics and to social unrest (particularly
in the form of civil war). So another moral emerges here, that it is hard to treat a
system in isolation. Virtually every system has to be considered as open to external
influences. This does not cause a great conceptual shift, but one nee ds to model the
probability distribution for the external influences and if they are themselves the
results of space-time phases for large complex systems this is not stra ightforward.
Epidemiology is a branch of complex systems science in which control is relatively
well developed. Governments have vaccination p olicies, movement reduction policies,
and identification policies for tracking down outbreaks of disease and limiting their
spread. These have been lear nt by bitter experience. Similar ideas apply to the spread
of computer viruses. An area in which there is need for more work is the spread of
ideas: some are deemed good, such as those has leading to reductio ns in smoking;
while others are considered bad, s uch as radicalisation (but this depends on the belief
system and who you ask). In either case, it is important to understand what makes
an idea spread or not.
Electricity distribution is moving into unknown territory with widely distributed
generation, often from highly variable sources like wind, and the consequent prob-
lems of bala nc ing supply and demand. The longterm solution is almost certainly a
real-time (and space) pricing signal, coupled to s mart consumption, generation and
storage devices which take or provide power when it is advantageous to them and
not otherw ise. How to design such a pricing sy stem to run stably is a major question.
Stability here does not mean that there would be no fluctuations; it mea ns that the
space-time phase would not have any large excursions.
Population, its geographical distribution, age structure and skill distribution is
an important issue. We have now pas sed seven billion peo ple worldwide and many
of the tensions in the world can b e a ttr ibuted to there being too many of us for
current technologies to cope with. Possibly this is a system which has not yet reached
a s pace-time phase but it may be adva ntageous to manage it onto one. For example,
a simple way to r educe family size is careers for women. Thus education leading to
24 Will be inserted by the editor
more women feeling they want a car e er is probably a good longterm solution. But to
model this in any serious way is a challenge.
Lastly, climate is the archetype of a spac e-time phase . Although there are crucial
aspects such as cloud for mation for which good models are not yet known, one can
hope to devise control strategies tha t move the climate in preferred dire ctions. The
main currently active control is CO
emissions and various geo-engineering co ntrols
have be en proposed.
There is clearly an enormous gap between these real world problems and the
nascent theory of manage ment o f complex sys tems sketched above. The biggest chal-
lenge is to develop models of social systems that capture the essence of human be-
10 Conclusion: Complexity Science in FuturICT
The FuturICT Flagship prog ramme is built on the three pillars of complexity science,
social science and ICT. In this paper we have laid out the main concerns and challenges
in the science of complex systems with special e mphasis on the Complex Systems route
to Social Sciences.
Although ther e is no agreement on a precise definition of the word ‘complex’,
there is wide consensus on the prop erties that can make systems complex. These
include them having: many heterogeneo us interacting parts; multiple scales ; compli-
cated transition laws; unexpected or unpredicted emergence; s ensitive dependence on
initial conditions; path-dependent dynamics; networked hierarchical connectivities;
interaction of autonomous agents; se lf organisation; non-equilibrium dynamics; com-
binatorial explosion; adaptivity to changing environments; co-e volving subsystems;
ill-defined boundaries; and multilevel dynamics. In this context, science is seen as
the process of abstracting the dynamics o f sys tems from data. This pre sents many
challenges including: da ta gathering by la rge-scale experiment, participatory sens-
ing and social computation, and managing huge distr ibuted dy namics and heteroge-
neous databases; moving from data to dynamical models, going beyond correlations
to cause-effect relationships , understanding the relationship between simple and com-
prehensive models with appropr iate choices o f variables, ensemble modeling and data
assimilation, and modeling systems of systems of systems with many levels between
micro and macro; and fo rmulating new approaches to prediction, fo recasting, and
risk, especially in systems that can reflect on and change their behaviour in response
to predictions, and systems whose apparently predictable behaviour is disrupted by
apparently unpredictable rare or ex treme events.
Undoubtedly great progress is be ing made, and European scientists are playing a
leading role in this fie ld. The ambitions of the FuturICT Flagship Project are high
indeed and huge advances in the science of complex systems will be necessary for
them to be achieved. ICT will continue to be at the heart of Complexity Science
and this science will generate many ne w ICT applicatio ns. Complex sy stems science
desperately needs to be better assimilated with social sc ie nc e and there are enormous
challenges and opportunities in this respect. Despite grea t progress, the science of
complex systems is s till in its infancy and we must not promise too much. This makes
FuturICT very high risk, but it is hard to see how humankind can face the future
without rapid advances in the science of complex systems.
MSM acknowledges financial supp ort for research on Complex Systems from MINECO
FIS2007- 60327. He also thanks Emilio Hernandez-Garcia for enlightening discussions
Will be inserted by the editor 25
on the subject. RSM is grateful to the Alfred P. Sloa n Foundation (New York ) for a
grant that is enabling him to address the research programme outlined in the section
on “C ontrol and management of complex global systems” We are all grateful to the
Future and Emerging Technology (FET) unit of the E uropean Commission for the
support it has given to developing and coordinating complex systems science over the
last dec ade.
1. P. W. Anderson, Science 177, (1972) 393-396
2. L. Pietronero, Complexity ideas from condensed matter and statistical physics, Euro-
physics News 39, (2008) 26–29
3. Managing Complexity: Insights, Concepts, Applications, D. Helbig ed. (Springer) 2008
4. L. Pietronero, Physicists get social, Nature Physics 6, (2010) 641-640
5. C. Castellano, S. Fortunato and V. Loreto, Statistical physics of social dynamics, Rev.
Mod. Phys., 81 (2009) 591–646
6. Ball, P., Critical Mass: How One Thing Leads to Another, (Farrar, Straus and Giroux,
London), 2004
7. Majorana, E., Scientia 36, (1942) 58 and Quant. Finance 5, (2005) 133
8. T. Schelling, Micromotives and Macrobehavior (Norton) 1978
9. R. Axelrod, J. Conflict Resolution 41, (1997) 203
10. D. Centola, J.C. Gonz´alez-Avella, V.M. Egu´ıluz and M. San Miguel, J. Conflict Reso-
lution 51, (2007) 905
11. J. H. Johnson, The future of the social sciences and humanities in the science of complex
systems, The European Journal of Social Science Research, 23(2), 2010, 115–134
12. D.Helbing, Science and Culture,76(2), 2010, 315; arXiv:1007.2818
13. Vicsek T, Czirok A, Ben- Jacob E, Cohen I, S hochet O, Novel type of phase transition
in a system of self-driven particles Phys. Rev. Lett. 75, (1995) 1226
14. C. Granger, Econometrica 37, (1969) 424-438
15. D.Y. Kenett , M. Tummin ello, A. Madi, G. Gur-Gershgoren, R. N. Mantegna and E.
Ben-Jacob, PLoS ONE 5(12): e15032 (2010)
16. S.Hempel, A.Koseska, J.Kurths, and Z. Nikoloski, Phys. Rev. L ett. 107, (2011) 054101
17. N. Wiener, Cybernetics: Or Control and Communication in the Animal and the Machine.
Paris, (Hermann and Cie), Camb. Mass. (MIT Press)
18. G. Soros, The New Paradigm for Financial Markets. Public Affairs, (2008), New York
19. Jerusalem Declaration on Data Access, Use and Dissemination, 2008
20. D. Lazer et al.: Computational Social Science, Science 323, (2009) 721-723
21. D. Helbing and S. Balietti, The European Physical Journal Special Topics 195, (2011)
22. Putting People on the Map: Protecting Confidentiality with Linked Social-Spatial Data, (Natl Acad. Sci., Washington DC, 2007)
23. D. Helbing, W. Yu, Proc. Natl. Acad. Sci. (PNAS) USA 107, (2010)5265-5266
24. M. B. Araujo and M. New, Trends in Ecology and Evolution 22, (2006) 42-47
25. J. M. Bates, C. W. Granger, Operat. Res. Qu art 20, (1969) 451-468
26. A. Gozolchiani, S. Havlin and K. Yamasaki, Phys. Rev. Lett. 107, (2011) 148501
27. E. Kalnay, Atmospheric modeling, data assimilation, and predictability (Cambridge Uni-
versity press, 2003)
28. D.J. Watts and S. H. Strogatz, Nature 393, (1998) 440
29. A. L. Barabasi and R. Albert, Science 286, (1999) 509
30. S. Bocaletti,V. Latora, Y. Moreno, M. Chavez and D. U. Hwang, Phys. Rep . 424, (2006)
31. A. Barrat, M. Barthelemy and A. Vespignani, Dynamical processes on complex networks
(Cambridge University press, 2008)
26 Will be inserted by the editor
32. R. Carvalho, L. Buzna, F. Bono, E. Guti´errez, W. Just and D. Arrowsmith, Phys. Rev.
E 80, (2009) 016106
33. S. Buldyrev , R. Parshani, G. Paul, H. E. Stanley and S. Havlim, Nature 464, (2010)
34. M. Gonz´alez, P. Lind and H. Herrmann, Phys. Rev. Lett. 96, (2006) 088702
35. Javier Borge-Holthoefer, Alejandro Rivero, Iigo Garca, Elisa Cauh, Alfredo Ferrer, Dario
Ferrer, David Francos, David Iiguez, Mara Pilar Prez, Gonzalo Ruiz, Francisco Sanz, Fermn
Serrano, Cristina Vias, Alfonso Tarancon and Yamir Moreno, PLoS ONE O NE 6(8): e23883
36. J. Leskovec and E. Horvitz, Planetary-scale views on a large instant-messaging network
(Proceeding of the 17th international conference on World Wide Web, ACM, 2008)
37. A. Arenas, A. D´ıaz-Guilera, J. Kurths, Y. Moreno adn C. Zhou, Phys. Rep. 469, (2008)
38. M. Zimmermann, V. M. Egu´ıluz and M. San Miguel, Phys. Rev. E 69, (2004) 065102
39. F. Vazquez, V. M. Egu´ıluz and M. San Miguel, Phys. Rev. Lett. 100, (2008) 108702
40. P. Holme and J. Saram¨aki, ArXiv 1108.1780 (2011), Physics Reports (in press 2012)
41. S. Havlin et al., Challenges of network science: Applications to infrastructures, climate,
social systems and economics in th is volume.
42. J.C. Gonz´alez-Avella, V.M. Egu´ıluz, M. Marsili, F. Vega-Redondo and M. San Miguel,
Threshold learning dynamics in social networks, PLoS ONE 6(5), e20207 (2011)
43. P.M. Johnson, Human centered in formation integration for the smart grid, Technical Re-
port, University of Hawaii, 2009. f
44. Al Gore, An inconvenient truth (2007)
45. Randers J. Meadows D. and Meadows D, Limits to Growth (Chelsea Green Publishing)
46. J. McGlade, Climate crisis needs empowered people, (2009)
47. Wang, Pu and Gonz´alez, Marta C. and Hidalgo, Cesar A. and Barabasi, Albert-Laszlo,
Understanding the Spreading Patterns of Mobile Phone Viruses, Science, 324, (2009) 1071–
48. Pickard, Galen and Pan, Wei and Rahwan, Iyad and Cebrian, Manuel and Crane, Riley
and Madan, Anmol and Pentland, Alex, Time-Critical Social Mobilization, Science, 334,
(2011) 509-512
49. E. Paulos, </