Socially-Inspired Information Technology

spongereasonInternet και Εφαρμογές Web

12 Νοε 2013 (πριν από 8 χρόνια και 3 μήνες)

484 εμφανίσεις

Inspired Information Technology

a simulation
experimental approach to the understanding and control of
complex IT systems using socially
inspired mechanisms



Social Systems are extremely complex and have many properties that wou
ld be desirable in IT systems. SIIT aims to
study and apply advanced socially
inspired computational mechanisms, in a similar way that biologically
computational mechanisms have been studied and applied. The methods will be the classic scientific

methods of
observation, hypothesis formation, formal modelling and experimental testing. The main tool, computational agent
based simulation, is used as an intermediate, descriptively
rich stage between the target complex systems and formal
models as wel
l as to explore, construct experiments and to try out ideas for applications. Thus SIIT aims to model the
complex workings of such mechanisms in a way which generalises beyond the particularity of specific examples, and so
results in more reliable applica

SIIT Proposal


B.1 Scientific and technical objectives and state of the art

Increasingly IT systems are connected to many others over large networks and interacting with other systems of diverse
designs and unknown purposes. Furthermore, people are increasingly i
nteracting with and through IT systems more
intimately than ever before. Where it used to be the case that interactions with IT systems could be analysed into a set
of separate, identifiable transactions, the interaction is now so rapid and tangled that I
T systems form

of the fabric
of society itself rather than merely being its
. The complexity of the social world and the world of IT systems are
becoming increasingly intertwined. We call such systems “complex IT systems” to distinguish them fro
m the paradigm
of unitary systems with a single designer and specified goals, where the users are external to the system.

The integration of heterogeneous, networked, IT systems requires new approaches for control and management in order
to ensure usabilit
y and reliability while not stifling innovation. The current approaches may either be; proprietary
systems, which do not inter
operate; unitary designs (e.g. ITU
T standards) which are very expensive and have fixed
functionality; and protocol component bas
ed (e.g. IETF standards) which are hard to manage and to understand by
users. In contrast to these; we are proposing to explore a model in which heterogeneous components adapt to each other
to provide robust, autonomic (easy to manage) functionality.

The S
IIT project aims at developing a new approach towards understanding and controlling complex IT systems, by
applying processes and mechanisms of self
organisation that have been shown to be valuable for natural societies

where self
organisation means the
spontaneous propagation of mutual influence and control rather than application of
centralised control. This new approach seeks to apply socially
inspired mechanisms that have promising features

features that will be especially useful in complex, unpredi
ctable and ill
defined environments.

The work of SIIT will contribute to the general understanding of complexity science, by creating systems with some of
the paradigmatic properties of complexity. Social systems with these features are:
: tasks in

the system are
distributed and duplicated, when needed there are parts to take over and share tasks.
: that the system capacity
increases rapidly with its size, usually because tasks are distributed without the necessity for expensive centralised
planning or global communication.
: the system can adapt to changing requirements and conditions, often by
changing its parts and their organisation dynamically.
: new resources and ways of dividing tasks can be
developed by the system fo
r extended functionality while taking advantage of existing infrastructure, i.e. the system can
‘grow’ as needed.
: the system is maintained via local interaction.
: the system is based upon
overlapping networks of parts which are endowed

with diverse properties.
: the system can not only change its
parts, but can shift from one part to another.

Current IT systems can have serious deficiencies in these respects, particularly if they are designed using a
comprehensive and fixed sys
tem model

under the false assumption that every condition and action of a system
component can be anticipated at the design stage by some sophisticated computation.

An IT system that is inspired by social interaction mechanisms consists of diverse parts
that communicate and
collaborate according to their predefined or emergent behaviours. These parts can be specialised in relatively simple
tasks and can form groups and develop value chains to solve complex problems. Furthermore, such systems often
e these properties without specific centralised planning and control (which is often brittle and inflexible).


systems have many abilities that would be desirable in an IT system. We have selected six

each of the
six is an
organisational ef

that we judge to be: achievable, modellable and applicable. This is not a complete list of
all possible such effects, but represent the proposed focuses of this project so that it is maximally effective.

Cooperative Group formation and maintenance
: Th
e ability to spontaneously form and maintain cooperative
groups even when this involves behaviour, which in the short
term is unselfish.

Creation of long value
added chains
: The ability to develop complex and long value
chains so that one part uses
the res
ults of other parts in a sort of ecology; this implies the ability to avoid destroying what other parts do.

Specialisation and complementarity
: The ability to organise so that different parts specialise in complementary
skills and tasks, allowing an effect
ive division of labour and the development of specialist skills.

Identification and mitigation of destructive agents
: The ability to isolate, counteract, hinder, or discriminate
destructive subparts. The ability to continue to operate even when many of i
ts parts fail or even act against the
rest of the system, often by simply learning to bypass these parts.

Constraining and structuring rules
: The ability to develop structuring and constraining rules to facilitate and
simplify interaction between its parts
. The ability to propagate such rules and get them applied by its parts, this
also implies the ability to identify, communicate, enforce, and possibly monitor the efficiency of the rules.

Collective innovation
: The ability to adapt in a decentralised and

flexible manner, by collectively learning and
propagating successful solutions among its parts. The ability to tell, acknowledge and predict emergence of
positive and negative interferences among subparts, and either neutralise or take advantage of them.

This project seeks to apply these properties of complex social systems

to the IT
systems that have to serve and operate in them.

SIIT Proposal


In this way SIIT hopes to facilitate the development of IT systems that are better fitted for the sort of messy
ts they have to operate in and serve the needs of society in a more natural and ultimately useful manner.

To achieve this SIIT has
two principal aims
, which may be characterised as scientific and engineering:

To understand the patterns of inter
action that occur in complex IT systems, through hypothesis,
experimentation and formal modelling using the tool of agent
based simulation.

: To apply this understanding to the production and self
organisation of complex IT systems
suitable for
operating robustly and adaptively in complex, rapidly changing and uncertain environments.

In our opinion these two aspects go necessarily “hand
hand”. Reliable engineering needs well
validated science to
guide and underpin it and good science needs re
al problems to inspire its development and ensure that it remains
relevant to its intended domain. Science without engineering can result in attractive theoretical structures and
techniques which, in practice, are almost impossible to apply. Engineering
without science means that one is locked
into particular techniques which have been found to work, but with little idea how robust this is to new conditions.

A major objective of SIIT is to
develop a methodology and a set of shared scientific norms
to enab
le the above aims to
be achieved. One way of characterising this is that SIIT aims to complement the relatively well
developed methods of
formal specification and implementation of IT systems (which has correct code as its focus) with those of experientia
validation (which focuses more on analysing and adapting the resulting behaviour). It is said that, even in closed IT
systems of a unitary design, 10% of the effort is expended on specification and implementation and 90% on debugging
and maintenance

IIT aims to develop methods that are applicable to the 90%, with respect to the complex, open and
fragmented IT systems that are rapidly becoming commonplace. This is not because of poor planning on the part of
system designers, but because of unavoidable

realities of software engineering (Parnass 1995). This methodology is
based on the classic scientific observational and experimental method and is described in greater detail in section B.4.

The principal aims are to be achieved via a sequence of sub
s with respect to
of the focus organisational
effects listed above. These are as follows:


To identify and abstract candidate mechanisms from (mostly human) society that have the potential for
facilitating the focus organisational effects.


To investig
ate/explore the workings of these mechanisms in complex simulations, particularly the extent to
which they facilitate the focus organisational effect.


To map
out the significant sets of environmental conditions for these mechanisms.


To check and compare di
fferent simulations of the same mechanisms.


To form precise hypotheses about the workings of these mechanisms with respect to sets of specific
environmental conditions, in particular when the mechanism may bring about the focus organisational effect.


To fo
rmalise these hypotheses in formal models.


To test these hypotheses in specifically designed simulation experiments.


To independently check these simulation experiments by other teams or persons.


To try the mechanisms against one or more canonical problems

relevant to the key organisational effect, so that
the formal models can be further checked and results from different mechanisms systematically compared.


To map out some of the scope of the formal models in different simulation conditions.


Where possible

to validate the results and the formal models by means of experiments to be carried out with
simulations of complex IT systems and in hybrid settings (with both humans and artificial systems).


To apply the mechanisms in demonstrators showing the potential

of applications and their reliability.


To attract and involve public and commercial bodies (especially SMEs) in the use of the techniques generated so
that the economy and community in general can benefit from the knowledge generated.

Of course, it is rar
e that such goals can be met in strict sequence and without a considerable amount of backtracking,
however such a set forms the principal ‘route’ for our method to progress along.

The long
term and speculative nature of this work means that it is inappro
priate for SMEs at this early stage (only one
partner, SICS, is an SME and that, is a research institute). It took many years before the ideas of Evolutionary
Computing started to be applied and, that we guess, we are looking at a similar time frame here.

Having said that, SIIT
does aim to involve SMEs particularly in the later stages of the project, when the commercial benefits are clearer and
the necessary investment in terms of development time is more reasonable for them. During the start and middle
of the project we will seek to ensure the maximum commercial involvement through our commercial partners, the
commercial associates of the other academic partners and the system of “client advisory boards”.

SIIT Proposal


This work will be supported by the parall
el development of: methodology; common standards; software tools; the
organisation and coordination of the project; the dissemination and training activities; and extensive consultation and
interaction with associated and other interested institutions. Th
e aims of these activities can be summarised as follows:

To develop a ‘catalogue’ of suggestions for good methodological practice in terms of this approach; to apply
these in the core work of the project and disseminate these widely.

To develop common stan
dards for the description of simulations, results, canonical problems and formal models
to facilitate the work in this field.

To develop a series of software tools to facilitate the practice developed, including: the production of semi
formal simulation de
scriptions; the development/transformation of these into a standard suitable for different
platforms; the execution of the code on different platforms; tools for analysis and exploration of results; etc.

To coordinate the work effectively so that: the work

progresses as smoothly as possible, any problems are dealt
with quickly; that the money is spent effectively and that knowledge is shared and reused.

To disseminate the results of the project as widely as possible in the EU research and commercial communi
and beyond so that the maximum benefit may be gained from it.

To train a new generation of researchers in the techniques developed in the project.

It should be clear that the nature of this work in terms of: the replication of simulations; the compari
son of results; the
coordination of the systematic mapping of results; the development of common standards of good practice, standards
and tools all require considerable
. This integration is not only necessary for the effective development of
this work, but is also
. Over the last decade the seeds of a distinctive European community of researchers has
developed, around the SIMSOC and MABS conferences which integrated social scientists who use simulation with A.I.
and (Multi) Agent System
s scientists, the ABS conferences, the SIMSOC mailing list, the journal JASSS, and the
Based Social Simulation group within AgentLink. Thus for the first time, there is a critical mass of experienced
European researchers available for an integrated p
roject. However, at the moment the work is disjointed

SIIT would
help integrate a core of such institutions and facilitate the emergence of a lasting and productive European field.

Thus SIIT also has the following, subsidiary aims concerning how the pro
ject is to be organised:

To further the social simulation field in the EU, by providing a core dedicated to integration and rigour.

To provide good practice, useful standards and software tools to researchers across Europe.

To be as open as possible in ter
ms of results, tools, workshops, training and decision making processes.

To involve as many other researchers as is possible and feasible in the work of SIIT.

There is now a short history of work that attempts to understand the workings of social mechanism
s in individual
and agent
based simulations. The most famous of these are Axtell and Epstein’s SugarScape model concerning
movement and the consumption of resources and Axelrod’s simulations on the evolution of cooperation in prisoner’s
dilemma game
s. This thread of work has resulted in the academic forums listed above. This work applies IT systems to
the understanding of social systems and has resulted in the community of agent
based simulators described above.

In parallel to this has been the dev
elopment of agent techniques and architectures as a method of IT engineering. This
is now a well
established field with its own major conference and journals. It has clearly taken some ideas from human
society and has applied them to computational system
s (roles, norms, trust etc.). It has concentrated upon formal design
and implementation, which has meant that it seeks to
the complexity of the IT systems it deals with so as to ensure
reliable system behaviour

this is explicitly spelled out in Ni
ck Jennings' guidelines (Jennings 2000).

A third strand has been the development of “complexity science” associated with the Santa Fe Institute. This approach
uses a lot of computational simulation and formal models adapted from physics to capture import
ant aspects of classes
of complex systems. This sort of work is now well
established and is a substantial component of the EXYSTENCE
NoE. It concentrates on how the interaction of large numbers of fairly simple and similar parts can result in complex
comes. Some of this has been applied to social systems. Although considerable progress has been made, there is no

theory of complexity (contrary to the hype that is sometimes given, mostly by those

in the field).

SIIT aims to do something dif
ferent to these strands. Its purpose is not only to use IT in understanding social
mechanisms but is seeks to ‘complete the loop’ by applying social understanding to IT systems in an analogous way
that biological mechanisms have been applied to IT systems
. Further it seeks to do this in a way that is complementary
to the approaches used in agent engineering. That is, it seeks to develop a methodology of analysis and adaptation of the
resulting behaviour of relatively complex computational systems, systems

where it is unlikely that careful design and
implementation will be successful
on its own
in ensuring that the outcomes are desirable. The kind of system that SIIT
focuses upon are composed of diverse parts, which frequently have unknown behaviours and g

this is in contrast
to the sort of systems usually considered in the fields of “complexity science” or “autonomous agents”. In particular it
includes mixed systems of human and computational parts, considered as a whole, and not just the software c
in isolation. In such messy and diverse systems it is unlikely that our intuitions will be so helpful in devising neat
style models, but will require more complex simulations as an intermediate step.

SIIT Proposal


B.2 Relevance to the objectives of th
e IST priority


in this section indicates a quote from the relevant objective]

SIIT has been designed to be

relevant to the objective set out in the FET proactive call on complex systems.

the objective is to create a new gene
ration of scale
free, autonomously evolving IT systems building on design and
control paradigms derived from complex system analysis

The particular approach of SIIT is to apply socially
inspired mechanisms to IT systems using a rigorous experimental
d, whose principal tool is agent
based simulation. Such systems will thus by their nature
incorporate adaptive
and stable self
regulatory mechanisms that guide their growth and lead to autonomous self

be able to
operate on multiple spatial a
nd temporal scales

continue operating reliably in dynamic environments
. The nature
of the IT systems that the developed techniques can be applied to is not, however, limited to societies of autonomous
based systems, but to

IT system which is

composed of parts that interact in ways that are not effectively
predictable (from outside the parts) and whose goals are not always known. This could include: distributed systems
with embedded controllers; ad hoc interactions across the Internet; and (v
ery importantly) mixed systems composed of
both software components and human actors.

To achieve this SIIT
will study real
world systems, in particular eco

and social systems
, within complex agent
simulations, in order to
understand how these scale
up and organise the information flow between their parts
. This
understanding is not only for its own sake but
in order to develop tools to “engineer emergent order”
. Whilst we
would argue that a
general conceptual framework for complex systems

is neithe
r useful nor possible (unless one has a
particular context that delineates what the relevant scope of ‘system’ is, as in physics), we do think that it is possible to

find hypotheses that generalise away from their original context and which can be captured

in formal (and hence
testable) models. This would enable
a leap from ad
hoc solutions to a scientifically
rooted paradigm shift
. Indeed the
paradigm shift we propose is away from approaches that might be characterised as grounded in
a priori

systems of
athematics, logic or philosophy towards an approach based on the classical scientific methods of observation and
experiment. The key to this is the use of complex simulations of an appropriate descriptive level as a stepping stone to
hypothesis and formal

modelling, since it is in the nature of the systems we are studying that the outcomes are often
intuitive and analytically intractable.

Thus, in particular, we aim to
create scale
free computational structures composed of self
assembling building
that are capable to develop

through spontaneous differentiation

organised structures and greater capabilities
Whilst we think that
languages for 'programming' such structures via local rules

are unlikely to be usefully captured in
a single comp
uter programming language, we do think that mechanisms for achieving known desirable organisational
effects can be developed that can be used as part of an engineering ‘tool
kit’ whose conditions of application are
indicated by their accompanying formal mo
dels. This ‘tool
kit’ of techniques does form the basis for a new language in
the widest sense, in the same way that a ‘vocabulary’ of architectural forms (arches, cross
beams, suspension etc.)
allows the ‘programming’ of complex buildings.

SIIT squarely
meets the objective of the IST future and emerging technologies programme. The interface between
computational social science and socially
related computer science
is an IST related field

with an accompanying
fledgling community
whose emergence would be g
reatly facilitated

by this project. SIIT clearly has
the potential to
become strategic for both economic and social development in the future
. It would obviously
feed in to mainstream IST
activities in the future
, particularly those on “Grid
based system
s for Complex Problem Solving”, “Communication,
computing and software technologies

Embedded systems”, “Knowledge and interface technologies

systems” and “Knowledge and interface technologies

based knowledge systems”.

It also meets

the overall IST objectives. It would
increase innovation and competitiveness in European businesses and

by providing a new approach to the production and control of the complex IT systems that it will have to
develop. Europe has developed a uniq
ue community of social simulation scientists with a very different approach from
those elsewhere, uniquely positioned to tackle the combined challenges posed by the increasing intertwining of IT and
society. The better meshing of IT and society that SIIT
will help facilitate would
contribute to greater benefits for all
European citizens
. In this way the potential impact of SIIT goes beyond a narrow technical objective of programming
an IT system better or characterising certain restricted classes of abstr
act system.

Finally SIIT has a role to play in achieving the Lisbon Objectives:
to become the most dynamic and most competitive
based economy

by helping unearth new techniques could be ideal in facilitating the dynamic organisation of
actors, inc
luding innovators, firms and consumers, in self
organising interactive webs allowing the interchange and
organisation of

and value. In a real way helping to import some of the dynamism inherent in human society
into the IT systems that interface

between actors in the economy.

SIIT Proposal


B.3 Potential impact

The impact of SIIT is, of course, uncertain, but potentially significant and diverse. Perhaps the most exciting possibility
is that since SIIT seeks to apply some of the distributed adaptive mechanisms

in society back to complex IT
embedded in

society, the result may be IT systems that integrate in a more natural, even human, way and hence
be more accessible and comprehensible to those who use them. One of the reasons for this is that the
knowledge and
techniques developed in SIIT will be applicable to systems composed of

of software and human parts. This is in
contrast to having to make big assumptions about the human needs in order to be able to design the IT component,
hoping that

the two will work well together once deployed. Another reason for this is that many of the mechanisms to
be explored in SIIT can be ‘bolted on’

the system is already running, thus it may be that using the techniques
developed in SIIT, whole systems

composed of people and computers can be

in an interactive way to bring
about desired effects

it is not necessary to redesign systems (or people’s way of using systems) from scratch. Such a
systems of post hoc adjustment may allow a process of
human and computer co
evolution to occur

a process that
might cause fewer shocks to the whole.

Other areas of potential impact include:


SIIT will result in formal models about the interactional dynamics of self
coordination and self
mechanisms, along with their conditions of application. Thus SIIT will have an impact upon the theory of some classes
of complex system, showing how empirically well
grounded theory can be developed that escapes the context of
particular simulations
. It will thus contribute to “complex systems” theory, or to be more precise, “very complex
systems” theory.


SITT will facilitate and support complex IT systems that are scaleable, robust, adaptive and self
organising, suitable for
deployment in

messy real
world electronic and social environments. It will bring forward new techniques for the
adaptive control of existing complex IT systems by importing regulatory mechanisms from society. Such techniques
could bring some level of control into sys
tems which have not been designed but simply accumulated, as well a systems
composed of inextricable mixtures of human and IT elements.


SIIT will result in the development of norms and standards for the specification of agent
based simulati
ons. Such
standardisation will facilitate the reimplementation of simulations, support cross
validation, and thereby greatly to
increase the rigour of the field. The development of code production and translation tools will be an important element
in th
is process which will result in the development of standards for the mark up and format of sets of results (including
a standard for the meta
data). Such a standardisation of results could make it much easier to develop an integrated suite
of analysis and

modelling tools with perhaps the automatic use of simulation models and results over the internet.


SIIT will create a core community of academics and commercial partners whose work is fundamentally integrated,
adhering to the common developed
standards, sharing tools, cross
validating each other’s simulations and building on
each other’s work. This would help cement the wider community of computer and social scientists in this area, setting
new standards of good practice and rigour, bringing t
he existing fledgling community to maturity.


SIIT will develop techniques to facilitate the rigorous reimplementation and comparison of agent
based simulation
models thus improving the methodology of simulation
based validation techniques to

up with the existing formal
verification techniques. It should also develop techniques for the large
scale mapping, exploration, analysis and
characterisation of large result sets, experience that should be increasingly applicable with the spread of


Software tools

SIIT will develop a suite of tools to support the good practice developed, including distributed agent
based simulation
frameworks and languages; tools for the conversion of standard simulation and result
set format
s, including suitable
training and documentation; tools to help document simulations and result sets, including the automatic recording of
their development history; and finally a set of interface tools to make the control of and results of simulations acc
to a wider audience. These tools will be useful to the wider community of researchers, and more likely to be taken up
due to the critical mass of SIIT partners who will already be using these tools.

SIIT Proposal


We also anticipate that the results of SIIT will

inform and help ‘leverage’ developments in GRID computing, embedded
systems and E
Science generally, as it will provide useful information about the possible interactive processes that
could occur between the nodes of such systems.


SIIT will

facilitate the sharing and dissemination of results to and from the wider community using a variety of media,
including: a web
site, training sessions, special issues, workshops and open
access database to promulgate theory, tools,
techniques and results.

Thus SIIT will have the effect of increasing the sharing and dissemination of knowledge in the
wider academic and commercial community. In particular its pioneering of a interactive portal and repository web
to coordinate and relate the knowledge c
reated might be reusable by other projects.

The tools that SIIT develops will be available to social scientists looking for quantitative methodologies and more
accessible, intuitive and indeed powerful modelling than they can get from econometrics and ga
me theory, which at
least currently only tell you about end states. Only agent
based models are good for ongoing, evolving social processes.


SIIT should result in IT systems that: can operate reliably within the complex and dynamic environment of r
societies; automatically adapt to changing circumstances; are less brittle to their set
up and use; can more naturally be
integrated and understood in social settings; and can facilitate social interaction among disparate parties with widely
goals. They could thus help embed IT systems more comfortably within society, making them more
transparent and, ultimately, more useful.


The technologies developed by SIIT for participating EU companies could stimulate a clutch of innovative prod
ucts and
services. To our knowledge, nowhere else in the world are there plans to exploit socially
inspired computational
mechanisms on the scale of this project. This alone will give a commercial advantage to firms in the EU. They could
also enable a mor
e general economic facilitation as a result of better business
business coordination in the EU and
elsewhere as a result of applying the techniques developed in this project to the design and coordination of, for
example: business
business communicat
ion systems, logistic systems, participatory web
sites and similar systems
and services.


A side
effect of SIIT might well be to throw up new ideas about the working of social mechanisms which might then
lead to new insight into how societies w
ork. Thus the results of SIIT might inform further investigation into the
workings of society, which may in turn inspire the investigation and application of further socially
mechanisms. Thus SIIT could help establish a positive feedback ‘virtuo
us circle’ from understanding of society and
complex IT systems and back again. This was the intention and, to a degree, the achievement of the ABSS SIG of


A similar side
effect may be on organisational science, by suggesting new w
ays in which organisations can be designed
or adapted. In the most general case we anticipate the discovery of new ways of organising other decentralised aspects
of society. In particular it may inform the governance of electronic institutions and inter
actions, using a mixture of the
rules and reputation mechanisms investigated.


Finally extrapolating from the planned project to the greatest extent, the work may inform thought about what ethics are
appropriate in new electronic business environment
s by ‘laying
bare’ some of the consequences of adopting certain
behaviours, attitudes or strategies. Significantly, it may be possible to show what kinds of ethical systems are stable or
‘workable’ in electronic organisations.

B.3.1 Contributions to stand

Wide application of agent
based simulation will depend more and more on the reusability, and the integrateability of
work produced by different groups. The development of new technology in the areas of agent
based simulation, and its
validation and ve
rification, therefore, immediately asks for the development of new standards.

The exact standards that will be developed will depend upon the experience of the partners in attempting to develop and
use them. Standards have to pass the test of usability a
nd utility before others will take them up

then they can be of
immense use in helping develop a field. Having said this, it is highly likely that standards will be developed for:

The semi
formal specification and description of complex agent
based simul
ations to facilitate the
reimplementation and checking of simulations.

SIIT Proposal


The formal specification of simulations for different sorts of platforms (declarative, GRID based etc.)

it is
envisaged that their might be a hierarchy of specification levels from
the generic semi
formal specification at the
top, though formal specifications that are suitable for types of platforms, down to formal specifications suitable
for implementation using a particular platform.

The appropriate meta
data for simulations and th
eir results, to facilitate making them available over the internet;
comparing them with other sets and the use of analysis tools.

for portals to (parts of) implemented agent
based simulations, enabling the development of a set of reusable
interfaces to s

and for verification and validation software.

Having good standards will facilitate cooperation between different groups in the EU that are interested in furthering the
rigour of agent
based simulation work. It will take an integrated project
like this to gather the critical mass and
common interest necessary for the development and use of such standards.

In a wider sense SIIT hopes to set new standards for the way in which social simulation work is done, by promoting
high standards of rigour

and openness.

SIIT Proposal


B.4 Outline implementation plan

The approach to be taken by SIIT has been chosen for suitability to the type of systems it will study. It is in the nature
of complex systems, and complex social systems in particular, that the behaviour of t
he system is not easily predictable.
The patterns and nature of the interactions between the parts of the system are frequently complex and counter
In IT terms, it is impractical to try to

the resultant system behaviour from its initial

specification and

there will almost always be surprising and significant divergences from the expected. Thus SIIT will
not concentrate on methods of formal specification and verification

these are already well
developed and are
ing to be developed by many researchers across the world. Although this project will
good practice in terms
of specification and implementation of systems, its
field of study
is designed to be complementary with these aspects.

The situation with comple
x social systems is different from that of complexity in theoretical physics, because, unlike
physics, there are no well
validated and reliable foundations upon which to build formal models. Unlike current
interesting attempts to apply statistical physics

and mathematics to modelling social dynamics (i.e. sociodynamics,
sociophysics), this project aims to describe at an appropriately level of analysis the phenomena produced by such
dynamics, and draw from them guiding principles, mechanisms and rules for t
he governance of IT systems. Not to
deny the utility of mathematical and physical models but that specific competencies are needed at the social and agent
scientific level in order to formulate assumptions, describe phenomena and formulate testable hypoth
eses about the
decentralised mechanisms that govern natural societies.

Rather the methodology SIIT will develop and apply is that of the
classic scientific approach
. Strong analogies can be
drawn with the methods used in ecology and experimental physics.

Ecologists observe and describe aspects of
ecologies in order to abstract what they hypothesise are important processes; they may then explore the ramifications
and properties of these processes in simulations and other formal models. SIIT will use the
observations and the models
of social structures and processes of social scientists, the models and theories developed within cognitive and
computational scientists to describe autonomous, flexible, adaptive, learning systems which interact, cooperate and
execute distributed tasks, as well as the descriptive simulations of social simulators in order to abstract what they
hypothesise are important mechanisms that might be crucial in the distributed self
organisation of society.

Physicists use a cycle of hypo
thesis formation, formalisation and experimental testing to discover the properties of
complex systems

SIIT will apply a similar methodology to complex computational systems. The exploration and
mapping of the detailed dynamics that result in the simula
tions suggests

about the sort of processes that
dominate in certain conditions and parameter ranges. These can be made precise as formal equation
based or
computational models of the subsets of the outcomes. Given the nature of the systems on
which the SIIT project will
focus on, we do not anticipate the development of formal theories of completely general scope. Rather one might expect
an incomplete patch
work of formal models to be more effective. It is important that the formulation of thes
e 'mini
models' is on the basis of the post
hoc exploration of (simulations of) processes rather than on unvalidated
a priori

theory (however formally or intellectually attractive). This is due to the frequently counter
intuitive nature of the
s that can result in these kinds of systems. It is already clear that such formal modelling will be possible for
some of the mechanisms proposed in this project and there are indications that it will be possible in the others as well.
This is not, of cou
rse, certain! This is the point of the exploratory projects in the core workpackages and the
fundamental reason why this project is suitable for the FET program

the results of the project are not assured, despite
the confidence and indications we have t
hat they will be forthcoming, but the potential benefits are great.

However partial these 'mini
models' may be, they will still be testable using carefully designed simulation experiments.
Just as with other experiments these need to be independently repl
icated, and can only provide certain disconfirmation
of the mini
model. Repeated survival of such mini
models applied and tested in different simulation experiments
indicates that the mini
model may be relied upon. The insistence of simulation replicatio
n is one of the novel features
of SIIT, this is onerous to do and, in the past, simulation builders have tended to avoid it. The integrated nature of SIIT
and its determination to develop and apply the best scientific practice means that for the first tim
e a core of
reimplemented and thoroughly investigated simulations will have been developed. It is hoped this core will encourage
others to aspire to the same standard.

The next step is to use the mini
model to help apply the mechanism to one of the projec
t's set of canonical test
problems, so that its performance can be compared (both in the degree of success and qualitatively) with that of other
mechanisms. The cross
comparison of proposed mechanisms against a shared 'library' of canonical problems is on
e of
the necessary integrative areas to be developed in SIIT. It is also a further test of the mini
models that they give
sufficiently accurate indications of outcomes in order to be useful in these, slightly more complex, test problems. Part
of the inte
gration in the project is the systematic comparison of different mechanisms against the same canonical test
problems. This is in sharp contrast to much past work in the area, which has tended to involve a new simulation and a
new problem in each investiga

The last stage is the development of, first, simulations of applications and then demonstrators of applications that put
the mechanisms into use (via their organisational effects). As with all such applications thorough testing and adaptation
is nec
essary in order to produce something that is useful. Here the mini
models can be used to make predictions about
the conditions necessary for establishing and maintaining the effects which gives a greater degree of reliability and
SIIT Proposal


confidence than would be
possible if the mini
model stage had been bypassed. An illustration of the relationship
between the entities dealt with is in












Tested in


guides and

gives idea of





ted in




leads to





. The major kinds of studied entitie
s in SIIT and some of the relations between them

Thus the core work of SIIT is to be organised into three broad (and sometimes overlapping) phases: the
phase, the
formal modelling
phase, and the
phase. These correspond to the work
as described above. To
summarise these:

In the
phase: the workings of
inspired mechanisms

are explored in
complex agent
, so the unfolding of the processes may be
, particularly with respect to
the extent to which they bring about one of the key
organisational effects

and (b) ‘mapping’ the
of the parameter/conditions where identifiable processes dominate.

In the
formal modelling
phase: the detailed understanding of the p
rocesses is collated. This forms the basis for
about the working of the processes within the
identified regions
. These hypotheses then form the
basis for
formal models

of the results exhibited by the simulations within these regions. These fo
rmal models
are then tested by specifically designed
simulation experiments
. Finally the
performance of the mechanisms
accuracy of their accompanying formal models
are to be

with others with respect to one or more
of the
canonical test p

In the

phase: the mechanisms are
to problems where the
corresponding organisational effect
is required. The formal models will be used to give approximations as to
the conditions under which one might
rely on the mechanisms
roducing these effects

this is a further test of the formal models. Finally
of the applications will be produced in order to involve commercial and other institutions to
develop the applications into useable form.

The core work of SIIT is
thus divided into:
formal modelling
projects. The exploratory
projects are characterised by the socially
inspired mechanisms that they investigate; the formal modelling projects focus
on a particular key organisational effect;

and the application projects on a specific problem or application domain. The
exploratory projects may feed into several different formal modelling projects, since the mechanisms they investigate
may play a part in brining about several of the key organi
sational effects. Likewise application projects may be related
to more than one formal modelling project, since they may require to bring about more than one of the key
organisational effects in order to work. Thus there is a complex (and not completely
known) set of dependencies
between the different projects in SIIT. The highly interrelated nature of the work in the core projects is one of the
strongest arguments for their
in a 6FP IP. A similar set of projects working on their own could n
ot be
expected to make as much scientific and engineering progress or achieve the standards of rigour that SIIT aspires to.

In order to structure the work, which is necessarily divided into a patch
work of interrelated projects, SIIT has been
organised t
o focus around 6
key organisational effects
. These are the six abstract properties of complex systems, the
key organisational effects
, listed in section B.1 above. The point of these is that each is potentially important in
applications (they represent a

desirable and applicable property) as well as being suitable targets for socially
mechanisms to aim to produce. Each of the core workpackages is focused around one of these key organisational
effects and will include: the mechanisms that may br
ing them about; the hypotheses and formal models concerning how
the mechanisms might act to bring them about; and the applications the effects are useful for. These core workpackages
form the ‘vertical’ packages of work in SIIT

the main ‘routes’ down wh
ich the three phases of work will progress.
There are, of course, substantial overlaps in terms of mechanisms, hypotheses, formal model type and application
SIIT Proposal


domain between the core workpackages, which is why it makes sense to integrate these core workpack
ages within a
single integrated project. An illustration of the structure of organisation of the core work of SIIT is shown in





























Demonstration of




Demonstration of




Demonstration of


Core Workpackage 2

Core Workpackage 1









. An illustration of the relation
of projects, core workpackages and phases

Each of the core workpackages has similar set of work plans, objectives, milestones and deliverables

although these
are adapted to the needs of their particular focus. The six core workpackages (and their shorte
r labels) are:

Core Workpackage on Group Formation and Maintenance (

Core Workpackage on Creation of Long Value
added Chains (

Core Workpackage on Specialisation and Complementarity (

Core Workpackage on Identification and Mitigation of Dest
ructive Agents (

Core Workpackage on Development of Constraining and Structuring Rules (

Core Workpackage on Collective Innovation: Learning from Observing Others (

These workpackages represent the key ‘hubs’ which enable the organisation of

the linking of the many mechanisms to
the many application domains. Some of the potential relations are mapped out in
. While this project will only
be able to investigate some of these, they it vividly illustrate the in
connectedness of this area as well as its potential.

SIIT Proposal


. An illustration of some of the
relations from mechanisms, through the six focus
organisational effects
, down to application domains (start from a mec
hanism, go right across to a
green rectangle, down to a red rectangle and then across to an application domain).

To support the development of the work in the core workpackages, there are a number of ‘horizontal’
. These touch on the w
ork in all the core workpackages, both in the sense of supporting the work as well
as helping disseminate the results. These five workpackages (and their shorter labels) are as follows:

Support Workpackage on Good practice (

Support Workpackage on St
andards (

Support Workpackage on Software tools (

Support Workpackage on Organisation (

Support Workpackage on Dissemination and Training (

A chart summarising the involvement of each of the partners in each workpackage is displayed in
. For clarity
some of the more complex cross
involvements between workpackages are not given, but the figure indicates the
principal commitments. For clarity these short (but hopefully memorable) labels will be used to ref
er to the
workpackages in this (Part B) of the proposal.

SIIT Proposal


. Principal involvement by partners in each workpackage

There follows an summary of each of the workpackages, with a special emphasis on how they integrate with eac
h other.
We start with the core workpackages. Due to their common approach they have many aims, methods, milestones and
deliverables with a common form. We thus start with these before going into each in detail.

Outline Implementation Features Common to

All Core Workpackages

Common Aims

To identify and abstract candidate mechanisms from (mostly human) society that have the potential for
facilitating the organisational effect.

To investigate/explore the workings of these mechanisms in complex simulations,

particularly the extent to
which they facilitate the organisational effect.

To map
out the significant sets of environmental conditions for these mechanisms.

To check and compare different simulations of the same mechanisms.

To form precise hypotheses abo
ut the workings of these mechanisms with respect to sets of specific
environmental conditions, in particular when the mechanism may bring about the organisational effect.

To formalise these hypotheses in formal models.

To test these hypotheses in specifica
lly designed simulation experiments.

To independently reimplement and check these simulation experiments by other teams or persons.

To try the mechanisms against one or more of canonical problems relevant to the organisational effect, so that
the formal mo
dels can be further checked and results from different mechanisms systematically compared.

To map out some of the scope of the formal models in different simulation conditions.

Where possible to validate the results and the formal models by means of experi
ments to be carried out with
simulations of complex IT systems and in hybrid settings (with both humans and artificial systems).

To apply the mechanisms in demonstrators showing the potential of applications and their reliability.

To attract and involve pu
blic and commercial bodies (especially SMEs) in the use of the techniques generated so
that the economy and community in general can benefit from the knowledge generated.

Common Milestones

Whether and to what extent the candidate mechanisms bring about the

desirable organisational effects.

SIIT Proposal


Whether and to what extent that significant parameter/condition ‘regions’ of the mechanism’s outcomes can be

Whether formal models can be developed that capture the outcomes in these ‘regions’, how accurate ar
e they,
and what aspects of the outcomes do they represent.

Whether the formal models (and the hypotheses they formalise) survive simulation experiments to test their

Whether the mechanisms and their accompanying formal models perform well again
st the canonical problems.

Whether the mechanisms and their accompanying formal models are usefully applicable to realistic problem

Common Deliverables

A set of simulations documented at semi
formal and formal levels, along with their code and a s
ummary of their
results (including reimplementations of simulations).

A ‘map’ of the significant parameter/condition regions for the explored mechanisms and the processes that
dominate these regions.

A handbook of formal models of the workings of the mecha
nisms in the most significant of these regions.
Including the conditions under which they hold, and the simulation experiments they were subjected to.

A set of simulation experiments documented at semi
formal and formal levels, along with their code and a

summary of their results (including reimplementations of simulations).

A set of simulations of applications documented at semi
formal and formal levels, along with their code and a
summary of their results (including reimplementations of simulations).

A s
et of application demonstrators.

A public summary report consisting of: case
studies following the process from exploratory to application stages
together with recommendations as to best practice.

General Relation of All Core Workpackages to Other Workpack

Core workpackages will be related to other workpackages in the following ways:

They may involve investigating some of the same socially
inspired mechanisms, albeit for different
organisational effects.

Their formal models may share similar forms and b
e based on similar techniques as those in other core
workpackages. Such cross
fertilisation is to be encouraged by joint project meetings.

They may be applied along with the results of other core workpackages in the same applications (indeed we
foresee th
e greatest potential for application where they can be combined).

The experience gained in the core workpackages will be compared, critiqued and collected in
, and good
practice learnt via the activities of

will aid and facilitate the progress and

rigour of the core work.

The standards developed in

will facilitate the communication and reimplementation of simulations and
results. Applying the standards will provide feedback to those developing the standards in

The software tools develop
ed in

will support the development of core work by supplying software tools.
Experience in using these software tools will inform their further development.

The knowledge gained in the core workpackages will be posted in the knowledge organisation we
site produced
as part of

so that it can be immediately related to the other knowledge. Experience in using this service
will inform its further development.

The training done in

will help researchers progress in terms of the core work more eff
iciently as well as
aiding the integration of the project by increasing mutual understanding.

The dissemination done in

will ensure that the results of the project are as widely accessible as possible.
Hopefully this will attract other researchers to

be involved in the work of the project in an unfunded way and
thus further the aims of the project

this is particularly true of the work in the first three support workpackages
, and
) where it is clear that as wide as possible involveme
nt and take
up is desirable.

Integration across Workpackages

The nature of the proposed work entails integration in many ways and at many levels. Work inside core workpackages
needs to integrated together in order for the maximum benefit to be gained in t
erms of both science and engineering
objectives. The work between core workpackages will benefit greatly from integration, in order to exploit the many
potential cross
applications of ideas and approaches. The support workpackages involve jointly develop
ing good
practice, standards, tools and knowledge coordination on a project wide basis (and beyond) so that this may feed back
SIIT Proposal


into the work in the core workpackages. Lastly the creation of a core community of researchers and knowledge,
dedicated to devel
oping and applying the best practice may help cement the fledgling wider EU community of computer
and social scientists in the field.

Ensuring relevance and quality

In addition to the normal methods of: meetings, objectives, reports, and deliverables. SIT
T will implement two specific
extra measures to help ensure the relevance and quality of the work done. The first is that of external client advisory
panels and the second is that of internal cross

We propose that each workpackage should have
a “panel” of two to four people external to the institutions in the project
who will review and advise on the direction and quality of the work. These people would be either from the
commercial sector with a remit to ensure the work remains relevant to th
e ultimate applications it is intended to be used
in and academics in the field who are not members of the project. The academics might well not be in the EU or
associated countries. This is immensely important for two reasons,

it greatly helps t
o have intimate access to an
outside view so as to maintain the highest quality of work and secondly it is a good way to involve people from the
commercial sector in the work of the project (and hence increase its impact). For new technologies like the on
will develop, a simple demonstration is often not enough

rather a greater involvement over a longer period of time is
often necessary in order for the potential to be realised. A specific amount of money has been included in the budgets
for payi
ng for the expenses (travel, hotel, meals) of such people

this is a necessary precondition for this kind of
involvement. The benefit to the work far outweighs the small additional cost in this regard.

, within the project it is incumbent upon th
e partners to reimplement and check the simulations of other
partners. This is rarely done because of the effort involved, but there is no better way of spreading deep understanding
of the mechanisms concerned. Too often have simulations gone unchecked,
if simulators are going to make a ‘step
change’ into a more rigorous science this is essential. This measure alone will have a large impact on the field.

Encouraging involvement by commercial institutions

Encouraging commercial involvement is essential to

the science as well as the engineering aspects of the work in SIIT.
Such involvement helps ensure the relevance of the more abstract work done. It is a crucial test of all successful
science that it does supply some additional ‘leverage’ upon the world.

The only way to test this is by attempting to
apply the theory to realistic applications. SIIT has four ways in which it will achieve such involvement.
through the commercial and semi
commercial partners in the project;

via the work o
f the partners with their
established commercial collaborators;
, via the commercial members of the “client panels” described above; and
through traditional demonstration and dissemination techniques.

Encouraging involvement by SMEs

The ‘blue

skies’ nature of SIIT, makes it less immediately attractive to SME’s. It might be some years before the first
fruits in terms of commercial applications is reached (although we hope that this will occur relatively soon). The
AgentLink road
map guessed a
t 2008 and beyond as the sort of timescale necessary for applications of this kind to be
realised. This is too long a time
scale for most SMEs’ investment horizons, they typically need a more immediate
return to any investment in employee time or other re
sources. Thus, in order to encourage their involvement, we plan a
targeted campaign in the last two years of the project. Representatives of suitable SMEs will be invited to attend project
workshops and visit partners. There is a small budget for travel
and subsistence costs that is particularly for this
purpose. It is thought that it is probably small consultancy firms, already involved, or with experience of, simulation
work or related IT systems that will be first interested.

Encouraging involvement b
y other researchers

Engaging/attracting the maximum involvement of other relevant researchers is crucial to achieving work of the highest
standard; for disseminating the results of the project as widely as possible and for getting the maximum value out of
project. To further this objective SIIT will advertise for other institutions to collaborate with SIIT as 'associate
members'. Such associate members do not enter into any contractual relation with the project, rather they are research
groups that wi
sh to be informed about and, possibly, contribute to the work of the project. These have been vetted
according to light criteria of relevance and quality. They will be able to: attend workshops and training events (at their
own expense); participate in the

work of the various workpackages (with the consent of those in the WP); receive
project electronic newsletters; and be able to access and contribute to the knowledge coordination web

It is expected that the overseas associate members already connect
ed with this project (see start of B.5) will apply to
their own national funding councils for money in order to be more actively involved with SIIT.

Policy regarding Intellectual Property

In all cases the 6FP rules have precedence in any matters concerning

intellectual property rights (IPR) and will form
part of the project contract and consortium agreement.

SIIT Proposal


Subject to the above rules and to the final negotiated project contract and consortium agreement, the following
subsidiary principles will apply.

All k
nowledge and techniques discovered or developed as a direct result of work in SIIT projects during the exploratory
and formal modelling phases of the project is to be considered as public knowledge to be disseminated and used by
society in general without
let or hindrance.

However it is recognised that in the application phases there may be developed techniques which are of commercial
value and of a kind to which IPR are applicable. If this looks like it will be the case any partner who intends to claim
full or partial rights over any knowledge or techniques must notify all partners of the nature of this claim as soon as
possible. In such cases legal agreement between all partners involved in the relevant development must be reached
before work can conti

In the case of any dispute the matter must be attempted to be resolved at a meeting of the management board on a
consensus basis, hearing all sides of the dispute.

B.4.1. Research, technological development and innovation activities

The research,
technological development and innovation activities are the prime responsibility of the core
workpackages. Below we describe each in turn.

Core Workpackage on Group Formation and Maintenance

Short Label:


Group formation is the phenomenon where different parts
of a system demonstrate the capability to cluster on the basis
of some shared or similar property. Such clustering can happen spontaneously, so that the phenomenon is not centrally
designed or coordinated. Group formation may happen in societies where shar
ed properties have either explicit or
implicit representations. In systems capable of spontaneous group formation, agents are able to construct categories that
partition the population and use these categories to influence interactions with each other.

up formation is important for a number of reasons.

It forms a basis for multi
agent cooperation. If agents need to cooperate in complex distributed systems, the
possibility for emergent group formation will greatly benefit agents with shared goals or inte
rests possibly
bringing different capabilities to the group.

It encourages coherence between agents in distributed systems, while it maintains diversity as groups are not
isolated but interact with each other.

Finally, within artificial societies, a grou
p is an essential concept that underlines the nature of the complex
dynamics of social structures.


To understand how the mechanisms of tags, trust and markets can facilitate the effect of cooperative group
formation and maintenance.

To form formal mod
els of the workings of these mechanisms of group formation and maintenance in such a way
that the mechanisms underlying group formation in complex systems can be applied in the design of new
complex systems.

Mechanisms to be investigated

: In complex a
daptive systems, tagging is a well
established mechanism for agents to use for identifying
similar agents. As such, agents have 'tags' to distinct them from each other and make them recognisable for
agents with similar tags. Such tags are normally imposed
on the agents by the system designer. Mechanisms
need to be discovered or constructed that allow for agents to generate their own internal representations of

: Trust is an issue receiving much attention in the multi
agent system field of re
search. For the dynamics of
group formation and maintenance, trust is a candidate mechanism or phenomenon that underlies group formation
in complex social structures.

: Here agents form groups on the basis of demand and supply.

Prospects for formal

modelling of outcomes

Work on group formation has mainly been empirical until now, while formalisation of discovered techniques and
mechanisms is upcoming. The formal representation of groups and organisations will benefit from developments in
formal veri
fication and validation investigations in the area of organisation dynamics modelling. As such, group
formation is the dynamic interaction in terms of agents joining and removing themselves from groups.

SIIT Proposal


Relevant canonical problems and tests: Relevant probl
ems and tests come from the areas of artificial life, biology, and
economics. In general, such problems are about individuals clumping together on a 2D grid with others of the same
kind; examples include the SugarScape test bed and the Schelling model. The

SugarScape test bed underlies many
researched problems and tests that are prototypical in the area of social simulations. Group formation has been
investigated in this test bed as to how similar resource traders find each other in a simulated gridworld. A

social science problem is the Schelling model of racial segregation, which essentially manifests as a game on a 2D grid
of locations of fixed dimensions, where locations can contain a black or white counter or be empty. Over time,
depending on s
ome set of fixed rules, we can observe self
organised segregation. An intended interpretation of
phenomena observed in the game is that even a desire for a small proportion of racially similar neighbours might lead to
organised segregation. Finally, T
he iterated prisoner's dilemma game models a common social dilemma in which
two players interact by selecting on of two choices, cooperate or defect; from four possible outcomes of the game,
payoffs are distributed among the individuals. Evolutionary group

formation is applied in the prisoner's dilemma game,
where the biasing of game interaction towards agents sharing identical tags is sufficient to produce high levels of

Relevant canonical problems and tests

Relevant problems and tests come fr
om the areas of artificial life, biology, and economics. In general, such problems
are about individuals clumping together on a 2D grid with others of the same kind; examples include the SugarScape
test bed and the Schelling model. The SugarScape test bed
underlies many researched problems and tests that are
prototypical in the area of social simulations. Group formation has been investigated in this test bed as to how similar
resource traders find each other in a simulated gridworld. A canonical social sci
ence problem is the Schelling model of
racial segregation, which essentially manifests as a game on a 2D grid of locations of fixed dimensions, where locations
can contain a black or white counter or be empty. Over time, depending on some set of fixed rule
s, we can observe self
organised segregation. An intended interpretation of phenomena observed in the game is that even a desire for a small
proportion of racially similar neighbours might lead to self
organised segregation. Finally, The iterated prisoner'
dilemma game models a common social dilemma in which two players interact by selecting one of two choices,
cooperate or defect; from four possible outcomes of the game, payoffs are distributed among the individuals.
Evolutionary group formation is applie
d in the prisoner's dilemma game, where the biasing of game interaction towards
agents sharing identical tags is sufficient to produce high levels of cooperation.


Environments or problems consisting of many interacting parts normally ask for g
roups of agents that can achieve goals
unobtainable for individual agents. Group formation is an organisational effect that needs to be controlled by identifying
its underlying mechanisms in order to support emergent group behaviour in such complex distrib
uted environments.
Examples include: cooperative efforts of many individuals in complex adaptive systems, information retrieval / partner
finding in shared
interest communities, trust
based negotiation in automated trading, agent
based self
ban traffic control, the formation of urban agglomerations, migration of distributed applications across a grid type

Relation to other workpackages


concerns the creation of long value added chains, the formation of which will often req
uire some form of
adaptive group formation. More specific the research in this workpackage will improve the study of the effect of various
types of networks depending on different group formation mechanisms on the robustness and performance of the

relates to this workpackage as both are concerned with forming relevant cooperation between multiple
agents. The work done within

will provide hypotheses concerning the construction of reputation systems as
systems of second
level discriminatory
cooperation among first
level cooperators (see the corresponding form).

will provide the exploratory basis that contributes to the theoretical framework of (desirable/undesirable effects of given
social norms implemented in different agent structures

Partners involved

The coordinator is VUA. Partners involved: in the exploratory phase BATH, Cemagref, TNO and VUA; in the formal
modelling phase MMU, TNO, and VUA; and in the application phase UCL and VUA.

Core Workpackage on Creation of Long Value
dded Chains

Short Label:


The dynamics of value
added chains (or networks) have been (and continue to be) studied in the fields of organisational
systems/IT and managerial/decision
making under various banners such as: 'strategic partnerships', 'interorganisational

systems' and 'electronic business integration'. It produces disintegration of the firms, the emergence of new markets, and
finally, the extension of the value chain. Sometimes the opposite effect happens, and if firms become more efficient
than markets, t
he resultant value chain becomes shorter.

Internet and logistics are physical systems affected and affecting the extension of value
chains. The supplier
relationships and the efficient firms’ organizational size are simultaneously cause and effect o
f the extension of value
SIIT Proposal


chains. The mechanisms that enable the emergence of new markets substituting some transactions internally done by
firms can provide a formal framework to reduce complexity in value


To understand the how the mechanisms

of: market mechanisms with transaction costs, limiting resources access
via property rights and diffusion of innovation may facilitate the creation of long value
added chains.

Mechanisms to be investigated

Economists explain such effect from two point of
view: Transaction Cost Theory (TCT) and Agency Theory.
When transactions are organised by the market, they imply some costs of information, negotiation and guaranty.
By other side, when firms internalise such transactions by the production of goods that co
uld be available in
external markets, they account for direct and indirect costs. The motivation would come if the costs of obtaining
a processed object were less than if the agent did the processing itself. This implies either a specialisation due to
ited skills (or the fact that skills are 'owned' and not commonly shared) or economies of scale. Of course,
where information is concerned one has the ultimate economy of scale because once some information has been
created (found or computed or learnt or

whatever), it can be duplicated at almost no cost.

Creating new markets is also not a merit in itself: several counter
intuitive phenomena regarding auctions, for
instance, have been detected, showing that these new markets do not function as theoreticall
y expected. It
suggests the exploration of two mechanisms:


The control of marketplaces

Why do companies with limited unique assets survive, although theory
points to that ordinary contracts between two individuals should replace them?


Alliances and the p
roblem of organizational boundary making

To which degree do companies that
build on technology which is global, instant and interactive, desire to define themselves?

Another view of the same thing is as a developing ecology. Any existing organism in a ni
che creates more niches
by its actions. New organisms can emerge to exploit these new opportunities etc. so that a 'food
chain' develops.
Thus fundamentally the mechanism is firm/individual/organism adaptation where the results of others are
available to
them for re
use (though maybe at a price), and there is some pressure for specialisation. This
pressure can either be a limitation in resources or ability. This is where Eric Baum's imposition of strict property
rights comes in

it forces all agents to
find somewhere where they can improve the efficiency of the whole

Prospects for formal modelling of outcomes

Dynamics of value
added "chains" are studied in economics, the new physics and more traditional statistical physics. A
new formal framework

can be well grounded simulating artificial societies’ behaviour. We consider that the costs
associated with transactions and production/transformation must be explicitly modelled. Hence, transaction cost theory,
needs critical examination, both in its ori
ginal form (due to Coase) and its revitalized form (due to Williamson).

The industry can be modelled as an artificial society of agents, where the supplier
customer, supplier
customer relationships and the institutional frame, enable the

extension or reduction of the value chain. In this
way, it is fundamental to formalise the role of intermediaries in information brokering markets, the enforcement of
property rights, to identify the key factors that apply in social firm
firm (firm’s n
etworks) relationships.

Relevant canonical problems and tests

A canonical problem in terms of distributed forward
chaining theorem proving, where developed lemmas are
then used by other agents in the proof of theorems, requires the construction of long val

Another canonical problem is that of collectively searching for information that a group of users with similar
needs want. Different agents can specialise in finding different kinds of information, which may then be
redistributed (or re
sold) t
o other agents who specialise in being brokers to others etc.

Potential Applications

To apply the understanding, simulations, algorithms and formal models to the following application areas:

The co
ordination in manufacturing environments, particularly to
solve job
shop scheduling problems. In
particular, we will develop a simulated factory to test under what conditions those mechanisms outperform
heuristics solutions for the scheduling problem and job
shop floor management.

The distributed searching of th
e web for a related set of information needs. Following (Baum 2002), sets of
crawling agents search the web for pages that match one of a set of users with different (but related) needs.
An internal market helps structure the feedback from the users
to the end
search agents via intermediaries who
act as information brokers. Specialisation is forced by the distributed and disorganised nature of the web

agent can only search a small portion of it. Complementarily is important for the efficiency

of the search
implemented by whole system of agents.

Web based P2P applications in which client / servers pass information and jobs between one
another in the form
of a “self
organised informational production line”. Such “production lines” may come into
being for a number
SIIT Proposal


of jobs and then dissipate if no longer needed. Consider for example a user wanting to enquire and book a
journey and accommodation that involved train, flight and hotel booking. A set of web agents (peer
client/servers) may form a chain

in which first the flights are booked then the train then the hotel. Each stage
applies constraints on the next

i.e. the train needs to be booked well after the plane is timetabled to land.

Relation to Other Workpackages

Value chains could be an aspec
t of the cooperation of tag
based group formation, both as a mechanism (i.e. extended
symbiotic relations) and an effect of the specialisation in groups, thus involving


mechanisms (evolution of competing institutions, proposal
and voting procedures) that can be also adapted by the
application projects.

Partners involved

The coordinator is UVA. Partners involved: in the exploratory phase SICS, Cemagref, and UVA; in the formal
modelling phase UniS, BT and UVA; and in the applicat
ion phase OFAI, UMIST, RWE.

Core Workpackage on Specialisation and Complementarity

Short Label:


is the property that parts of a system only concentrate on a restricted range of tasks or skills.
is the property that the specialisms o
f the parts are complementary to each other and not all
concentrated on the same set of tasks or skills. The presence of these properties mean that different parts of a system
should specialise in different aspects or stages in a problem or environment wit
hout this being coordinated by a central
designer or planning module.

These are important for several reasons:

in distributed systems one does not want too much duplication of effort;

being able to specialise allows for the easier development of either s
impler or more sophisticated problem
solving (or computation) by the part;

it encourages all aspects of a problem to be addressed to some degree

not just the easiest aspects;

it encourages diversity in problem solving approaches, which can give the who
le system robustness against
unexpected changes, since solutions currently in a minority can seed a new majority.


To explore and understand some ways in which specialisation and complementarity can be brought about