The Micro-Macro Link in DAI and Sociology
Michael Schillo1, Klaus Fischer2, Christof T. Klein3
1Multi-Agent Systems Group, Saarland University, Im Stadtwald,
66123 Saarbrücken, Germany, email@example.com
2DFKI GmbH, Stuhlsatzenweg 3, 66123 Saarbrücken, firstname.lastname@example.org
3Department of Sociology, Saarland University, Im Stadtwald,
66123 Saarbrücken, Germany, email@example.com
Abstract: No matter if a population is human or artificial, we can surely
identify phenomena that can be described as micro or macro phenomena. In this
paper, we discuss micro and macro aspects of a population from a DAI and a
sociological point of view. We analyse similarities and differences in these
viewpoints, and identify misperceptions in the DAI community about the
micro-macro terminology. We explain these misperceptions and argue for the
transfer of sociologically founded concepts to agent-based social simulation.
Our research is done in the DFG focus programme socionics. We cooperate
with sociologists from University Hamburg-Harburg with the intention to
transfer knowledge from sociology to DAI as well as from DAI to sociology. In
cooperation with DFKI Saarbrücken we work on improving agent theories to be
applied in large sized multi-agent systems in the freight logistics domain.*
The problem of how individual action and structural rules in a set of agents interact is
a foundational issue for both DAI and sociology, also known as the micro-macro
problem. The understanding of the link between micro and macro would mean a
substantial advance in designing agents for dynamic and large-scale agent-based
social simulation, as well as a deeper understanding of human societies. Furthermore,
modelling the macro aspect in agent theories is considered to be essential for DAI
research, as this concept substantially contributes to the distinction between artificial
intelligence and distributed artificial intelligence (DAI). For this enterprise, a
scientific cooperation with sociology can be of great benefit to DAI. However, we
found that a mutually agreed terminology cannot be assumed.
The micro-macro problem is perceived in distributed artificial intelligence (DAI)
research as a central issue because it directly refers to such problems as coordination
and scalability. And indeed, the definition of distributed AI as opposed to the parent
discipline of artificial intelligence heavily depends on aspects that are only introduced
by the problems that occur when multiple actors face the results of each other's
actions . Not surprisingly, there are differences of definitions of the micro-macro
problem as researchers perceive it in the DAI community and the perspective taken in
mainstream sociology. DAI definitions of the macro level either intend to abstract
* This work is supported by Deutsche Forschungsgemeinschaft under contract Fi 420/1-1.
from the individual and to summarise certain features of a group of agents
(performance, communication overhead etc.) or aim at mechanism and organisational
design. While the former is a descriptive approach, the later is normative. Sociology
does study the same level of abstraction from the individual, but takes a different (and
usually only descriptive) perspective. In sociology the macro level of a society is
itself a structure, which possesses to a certain degree it's own autonomy: it survives
the individual and is (primarily) independent from the influence of any single
individual. A further important feature of the macro level is that it reproduces itself
over and over again by channelling the interests of the individuals.
While organisational theory by definition does not make any claims about how a
society (including a number of organisations) is composed, reducing the complexity
of a society to a multi-dimensional performance vector does not pay tribute to the
complex dynamics that can be observed at the macro level of human societies. This
does not render the cooperation of the two fields obsolete. On the contrary, looking
further at sociological theory is most fruitful to DAI research. Apart from the solely
action-oriented or structure-oriented theories, there is a selection of hybrid theories
that try to explain the connection between individual action and social structure
(Giddens, Bourdieu etc.).
In the discussion section we propose and start to analyse the habitus-field theory of
Pierre Bourdieu, which tries to explain the effect of individual behaviour on societal
structures and vice versa. This is where the great strength of the theory lies and where
we expect that DAI will find a lot of concepts for overcoming the micro/macro gap.
For example we state that the theory on this reciprocal relationship is the medium that
answers Castelfranchi and Conte's  question of how cognition can be structured by
society and what is essential for the emergence of structure from micro-interactions.
Our research is done in the context of the field of socionics . In this area
sociologists and computer scientists try to transfer methods and theories from one
discipline to the other. Our main concern is the modelling of interactions in the
domain of shipping companies. This scenario is defined by its openness and
complexity as we encounter a great diversity of agents as well as tasks and time
restrictions. Typically is also the large scale of such MAS in the magnitude of
thousand agents that requires not only interaction on a micro level but also macro
structures to function efficiently, and coherently. Our work leads us to the conclusion
that building social agent architectures that can deal with both, micro and macro
phenomena is not solely for the purpose of human adequacy but has also strict
engineering reasons. This emphasises the importance of sociologically founded
theories applied in DAI research.
2 The Micro Level in DAI and Sociology
The micro level is the area where we can expect to find mostly agreement between the
two disciplines. The micro level is composed of individual actors (humans and agents,
respectively) that interact. However, both disciplines emphasise different aspects.
DAI focuses on the cognitive architecture and the theory of how to model knowledge
acquisition and memory, perception and problem solving. This results in a focus on
designing algorithms that produce for a given input an appropriate (rational?) output,
as expressed by the widespread acceptance of decision and game theory. Sociology on
the micro level however, focuses on interaction and relationships between actions and
actors. Also, sociologists consider social actions, i.e. actions that are aimed at
changing the actions, effects of actions or beliefs of another individual. It is important
to note that this excludes actions like unwillingly causing an effect on another person
and actions aimed at objects, but includes actions like threatening another person
(social in the sense of related to other individuals and not in the sense of caring).
These differences may seem subtle at first. As we go on to take a look at the macro
level, this differences become more important, as the perception of what is interacting
on the macro level diverge significantly.
3 Overview on Perspectives on the Micro-Macro Link in Sociology
An exhaustive discussion of the definition of the macro concept in sociology
definitely exceeds the space provided here (and our competence). In fact, this
discussion fills volumes and some will even argue that this discussion is equivalent to
doing sociological research. We can note that many definitions exist, all tailored to a
specific theory, with no apparent success in the discipline to generalise from specific
theories. A second problem with presenting a clear-cut definition of the sociological
notion of the macro level may be that there is no corresponding physical fact in
reality. Even the phenomena usually connected with certain levels (e.g. interaction for
the micro level) are hard to pin down as they sometimes are used with slightly
differing connotations (e.g. when talking about the interaction of religion and politics
as their bi-directional influences, which are phenomena of the macro level).
Depending upon perspective of observation, the subject of social sciences can be
examined thereafter similarly from micro, meso, to macro or metasociological
perspective. The missing of a generally accepted theory of the social leads to
distortions and formation of different schools with according to differentiated research
programs. Thus different paradigms co-exist for the study of the emergence of social
structure in contrast to Kuhn's thesis on „changes of paradigms” . A brief
description of the four perspectives follows:
Micro-level: Sociology as science of social concern and interhuman behaviour.
Investigation of the influences of small groups on the non-standard behaviour
(concern, perception and thinking), e.g. groups and exchange theories.
Meso-level: Sociology as science of the social institutions and organisations.
Investigation of the influences from social organisations, e.g.: organisation sociology,
work sociology, technique sociology, sociology of education.
Macro-level: Sociology as science of the whole society, its stability (static aspects)
and change (dynamical aspects). It analyses which forces are responsible for stability
and change: religion, economics, culture, institutions etc. Investigates the influences
of the 'society' and culture, e.g. general system theory, sociology of culture.
Meta-level: Sociology as science of the ideas about society and as criticism of
ideology. Investigates society and culture constructing ideas, objects and values, e.g.
knowledge sociology, social philosophy, critical society theory (Frankfurter Schule).
(Skinner)Individual institutions/organisations social systems/culture
Micro Meso Macro
Fig.1 Overview on foundational strands in sociological theories.
For example the introduction of a bank holiday will surely provoke a wide discussion
in a society. Politicians, trade unions, employer associations, even the churches will
engage in a debate on the advantages and disadvantages, all parties with their
respective motives. Observing the influences of the different fields that interact here
(economics, politics, religion) is observation of the society on the macro level. A
meso level view would be e.g. the investigation of the different groups involved.
Maybe a new movement will form that aims to prevent this bank holiday. An
investigation of this movement would be a meso study. A micro level observation
would be, if we looked at individuals in a group confronted with this topic and how
they interact, which group processes exist that shape the interaction etc. A meta
observation could be how eastern societies and western societies differ in decision
making on topics that involve economics, religion and politics.
When evaluating current sociological theory, we need to take into account the
classical theoretical works in this discipline. The literature on social theory presents
itself as complex and multi-layered. The social life as the shared object of
investigation was re-built as a complex variation of phenomena, depending on
observation levels by the examining scientists and their specific ways of examination.
In order to classify the parts of the social universe, it was broken up into four levels as
In a short overview we will present a collection of social theories and briefly
discuss their ranges and main features from the perspective of DAI. We divide the
broad range of theories into theories focusing on the micro level, focusing on the
macro level and theories that try to translate from the micro to the macro level and
back (see Figure 1). As we focus on the micro-macro link, we leave the more abstract
meta level out of scope of this discussion. Of course, this overview is reducing the
theories to an absolute minimum and will by many (sociologists) be viewed as lacking
respect for the complexity of the theories. However, this overview is not intended to
cover the theories in their details, this would be impossible in the space given to a
paper and a complete meta-analysis is left to scientists with more competence.
3.1 The Macro-Approaches
In the centre of these approaches lie large social formations or collective processes
(the objective structure). Their objects are for example the structure and the change of
governmental organisations and institutions (e.g. capitalist society formations as
strata, classes, parties). The main interest is to attempt the analysis of the whole
society by its objectified social structures. The aim of this macro-orientation on social
life phenomena is to describe and explain processes of reproduction (static aspect) and
social change (dynamic aspects) of societies under economical, social and cultural
points of view. The society is to be considered as an reality of its own, which can not
be deduced from individual contexts (i.e.. from acting and behaviour). In this view the
society does not comply with the sum of its parts.
The individual subjects play a minor part for the constitution of social life and its
actual conditions. In fact, by reconstructing the social in macro-models their influence
on social structure merely occurs as exchangeable data (contingent functions of
individuals). See for example the so called normative paradigm of Parsons' action-
theory : Confronted with social expectations (may-, shall-, must-expectations),
the owner of a social position (objective social structure) will, in spite of Parsons
voluntarist assumptions, be forced under societal conditions with different degrees of
sanctions to adapt to the objective structure. Thus, the homo sociologicus is viewed as
fulfilling obediently the integrative and forced upon function of the more abstract
layer in the social system. To give an example: In Luhmann's conception of social
systems  the actors were completely excluded with the definition of
communication as the basic element of modern societies and the selective process of
information, mediation and understanding, his theory of social systems defines the
individual (psychical systems) as environment to the system, which can only
participate to the social by communication. (cf. [36, 38, 1, 2, 46]).
3.2 The Micro Approaches
The micro-sociological approaches study the social by observing the individuals and
their interaction behaviour (e.g. ). The issue most important in this research area
is: How can individual behaviour (action, mind, cognition) with no explicit and
planned coordination create the social, i.e. the emerging of social coordination and the
given structures1. The dependence on the social structure, surrounding the individuals
is not rejected, but plays a minor part in this perspective. As a reaction and critique on
the objective (i.e. macro-) perspective and its assumptions of a social organism, of
functional adaptation of individuals to the system in the first half of the century, the
1 The following references give an overview to these approaches: The phenomenological
approaches in succesion of Alfred Schütz, for instance  or . For the symbolic
interactionism see [32, 5, 18, 13]. For the utilitaristic/behaviourist paradigm see [22, 9, 19].
micro-perspectives received increased attention (see the critiques on Parsons by
Schütz, Mead, Blumer etc.). A second motivation was the intention to reduce the
scope of society analysis to the social psychology scope of learned behaviour and
exchange processes in group theory .
Micro-perspective approaches try to investigate how humans typically act under
the assumption of the presence of the generalised other (see Mead's concept of
identity as intersubjectivity and human gestures as significant symbols). These
approaches pose the question of which motives and expectations guide the
individual's behaviour. They try to reconstruct these motives and expectations from
observed situational contexts and behaviour (see the interpretative paradigm, a notion
which summarises the approaches of Schütz and Mead as well as their followers).
3.3 The Hybrid Approaches
The opposition between micro and macro-approaches belongs to the classical debates
of the sociological community. But besides the traditional antonyms corresponding to
the micro-macro clash as for example subjectivism-objectivism, system theory vs.
theory of action, collectivism vs. individualism etc., we have to note a „renaissance„
of the question about the relation of society (structural aspect) and the individual
(action or cognition aspect). The main target of the „hybrid movement„‚ was to
explain social life in relation to both action and the structure, like for example
Anthony Giddens did .
One of the sociologists with great importance in this respect, not only in France,
but all over the world, is Pierre Bourdieu. The conceptualisation of the habitus
concept (first 1967) allowed Bourdieu to develop the dialectic relation of objective
structure and subjective action/cognition by the assumptions of internalising the
structure and reproducing social structure in individual life styles, according to the
position in the social space. In contrast to Giddens who created his concept of
structuration structure for theoretical reasons, the habitus was created and based on
Bourdieu's practical work in ethnographic field research in North Africa .
4 A brief Summary of the Notion of the Macro-Level in Sociology
Viewing the macro-level of a society means to attribute autonomy to the structural
aspects of a social context. These aspects cause stability and change and can be
summarised by such concepts as religion, economics, culture, institutions etc.
Autonomy here means that no individual does have the power to change these
structures and it will even be difficult for a group of individuals. It also means that the
structure is not dependent on the existence of a specific individual, the structure
survives the individual. While this independence of structure from a specific
individual holds, it is also true that the structure depends on the whole population for
reproduction of the structure (where reproduction is the only aspect the individuals
can influence). It is important to note that this reproduction happens even without
explicit knowledge of the individuals. The dynamic that exists in any given social
structure is created by the malallocation of resources to individuals. The structures
that develop are created as means of reduction of the complexity of life. In this sense
society or organisation can only exist if and only if participation of the individual is
the „reasonable“ thing to do.
Fig. 2 The relationship of structure, conflict and learning from a socionics perspective .
In this context it is interesting to remark the connection between structure and
learning (or adaptation) on the individual level (see Figure 2). Learning is a cause for
structural changes (changing goals, needs and ways of the reproduction of structure)
and structure shapes the rules that constrains what and how the individual can learn.
This is a connecting point to the idea of the construction of intelligence from the
societal context . But there is a second (indirect) connection via the concept of
conflict: conflicts are stimuli for learning (e.g. reinforcement learning) and learning
may lead to conflicts. Conflict again is connected to structure, as the change of
structure often leads to conflicts and conflicts tend to be the causes for such structural
5 In Contrast: The Macro Level in DAI
Firstly, we will look at the trends in sociologically motivated agent-based simulation
and will give a brief survey of the different applications of the micro-macro
distinctions. Secondly, we will look at what can be called application-oriented multi-
agent systems. Conte and Moss  divide social simulation (not DAI) roughly into
these two approaches and we will adopt their terms. The first (sociologically
motivated) set of research seeks to develop the foundations of social theory by using
DAI in theory testing by simulation, which Conte and Moss call the foundational
approach. The other approach, which they name the representational approach,
develops modelling techniques and agent specifications to represent observed social
and institutional processes. The first set of models and implementations can be
viewed as being primarily object to knowledge transfer from DAI to social sciences,
whereas the second set may benefit from sociological knowledge in terms of better
5.1 Agent-Based Social Simulation: The Foundational Approaches
Firstly, there is social simulation research that is inspired by game theoretic
approaches, which for instance includes the works that build on Axelrod's research
. These works concentrate on modelling attitudes (altruism vs. egoism,
benevolence vs. individual rationality) and improve these notions e.g. by mechanisms
for protecting cooperative agents from self-interested agents . These works can be
viewed as looking at the micro, i.e. interaction level of societies.
A more behaviourist strand of research is the work on platforms like
SUGARSCAPE and SWARM (e.g. ). Here the macro level is perceived as
patterns that emerge from simple behaviours in large sized populations. However, this
cannot be attributed as social actions as in these models there is no notion of self and
others and no action that is intended to influence another individual's belief or actions,
which is the very prerequisite of social action2. A definite exception in this strand of
research, are anthropological models that try to elicit emergent structures from social
behaviour (behaviour that is directed at other individuals). An example of such
research is the EOS project , which can demonstrate the emergence of in-group
hierarchies, which in sociological terms is a meso-level feature (as the relations of
groups are the subject of study).
In the previous approaches the macro-level is perceived as the overall behaviour of
a population of agents, an emergent structuring that is not hard-wired by the designer.
This is different from the sociological point of view in the respect that sociology
would require a number of hierarchies and groups to form, interact and cause changes
bi-directionally between micro and macro level.
There is also a strand of research that tries to explicitly model macro structure of a
society. However, such multi-level social simulation does not necessarily imply the
full bandwidth of sociological concepts of societal levels. For some good reasons
(modelling effort, simulation speed) it is common practice to restrict the simulation to
only a uni-directional relationship between micro and macro level, which still render
impressive results. E.g. Troitzsch  describes a multi-level simulation where
individual (behaviour) was simulated to make predictions about money spending
behaviour of a population, attitude formation in a population with no structural
changes, gender desegregation in schools etc. In these kinds of simulations the macro
level information consists of an aggregation of the micro level data. The design rules
out any possibility for the individual to change the structural constraints imposed on
the population. According to Conte and Castelfranchi  the preference of the uni-
directional link for social simulation in current research does not only hold for the
micro-to-macro direction but also for the reverse.
5.2 Application-Oriented Multi-Agent Systems: The Representational
According to Weiß , the micro-macro problem poses a question, which raises the
issues that define the term of DAI research itself. Therefore we will revisit these
issues, before we try to make out important strands of current research and how they
relate to the micro-macro discussion. It is important to note that although the micro-
macro problem plays such a central role, it is not a standard term in the literature (e.g.
2 However, we note that in agent research it is now a common understanding that social ability
for an agent does not only mean that the agent can communicate via an agent communication
language, but it also implies that the agent is able to model itself and others, reason about
when to communicate with whom, about what and in which way.
). In most of the literature it is referred to only implicitly by trying to decompose
the problem into several subproblems.
The first influential collection of such subproblems where we can study at least the
implicit notions in DAI of the micro-macro problem is the book by Bond and Gasser
. They list five central issues for DAI:
• How to enable agents do decompose their goals and tasks, to allocate sub-
goals and sub-tasks to other agents, and to synthesise partial results and
• How to enable agents to communicate. What communication languages and
protocols to use.
• How to enable agents to represent and reason about the actions, plans, and
knowledge of other agents in order to appropriately interact with them.
• How to enable agents to represent and reason about the state of their
interaction processes. How to enable them to find out whether they have
achieved progress in their coordination efforts, and how to enable them to
improve the state of their coordination and to act coherently.
• How to enable agents to recognise and reconcile disparate viewpoints and
conflicts. How to synthesise views and results.
Please note that compared to the sociological notion of the macro level, these issues
are more dealing with agent interaction than societal issues. Moulin and Chaib-Draa
 add a software engineering (or normative) perspective to this perception of DAI:
• How to engineer and constrain practical multi-agent systems. How to design
technology platforms and development methodologies for DAI.
Jennings, Sycara and Wooldridge  focus on the coordination aspects in DAI when
• How to effectively balance local computation and communication.
They approach the macro-level from a pragmatic point of view when formulating the
last issue for DAI:
• How to avoid or mitigate harmful (e.g., chaotic or oscillatory) overall system
This issue is also addressed by a range of game-theory-inspired research, usually
summarised under the term mechanism design (e.g. ). Weiß reformulates these
last two issues into the following desiderata:
• How to enable agents to negotiate and contract. What negotiation and contract
protocols they use.
• How to formally describe multi-agent systems and the interactions among
agents. How to make sure that they are correctly specified.
• How to realise „intelligent” processes such as problem solving, planning
decision making, and learning in multi-agent contexts. How to enable agents
to collectively carry out such processes in a coherent way.
Especially the last notion seems to be central in DAI: The design of agents that
behave coherently. This notion reflects the system designer perspective of a MAS and
for the application of MAS we assume that this notion is a cornerstone of the
perception of the macro concept. Only occasionally the macro concept is made as
explicit in the DAI literature as by Nwana :
„macro issues, such as the interaction and communication between
agents, the decomposition and distribution of tasks, coordination and
cooperation, conflict resolution via negotiation, etc. [The goal of macro
research] was to specify, analyse, design and integrate systems comprising
of multiple collaborative agents.“
Please note that none of the listed issues deals with the features required by sociology
for societies, e.g. power, institutions etc. The term conflict only occurs in the efforts
to avoid it (this is the aim of work on coordination and conflict resolution) and
although there is a tremendous concern for patterns of actions, until now there seems
to be no theoretical framework to formally analyse such patterns. Verhagen and Smit
 attribute this to the different approaches of sociology and DAI, where DAI (by its
continuingly strong connection to the cognitive sciences) is more concerned about
action selection and cognition than the limitations imposed by societal structure on
the individual and the effects of knowing about these limitations. Although there is
some work on recognising and reasoning about relationships, namely goal/task
dependence  and role definition and role dependence , we cannot say that they
approach the far more complex forces that are active on the macro level. Rather, these
theories cover the group or organisational level of society.
The confusion of the macro concept between sociology and DAI is partially due to
the fact that there is also a (minor) perception of macro as being the structures and
rules on the top level of the social context as it is perceived by the individual. In this
sense, any given simulated population will have a macro level modelled as well. This
holds for prehistoric human communities as well as for even the simplest community
model in DAI. However, the majority of social scientists views the most complex
level of today’s human society as the measure with which the macro level of a popu-
lation ought to be analysed.
6 Four Misperceptions in DAI Research about Social Phenomena
In this section we apply the sociological notion of the micro and macro level of
society for a discussion of the use of these metaphors in DAI research.
1) Mechanism design is macro-level design
Mechanism design is usually the coordination of actions of individuals to achieve
some invariants of the behaviour of a group of individuals (; etc.). However,
unless there is structure or dynamics in the system that goes beyond the single
interaction, there will be no manifestation of societal structures or institutions. In
social psychology there is a collection of work inspired by game theory on penalty
systems and their emergence in games (e.g. ). This could be viewed as advancing
to the meso (group) level. Modelling processes among individuals is to be located at
the sociological micro level
2) Macro-level behaviour is emergent behaviour
According to Langton  emergence is a „result that was not defined statically„ (i.e.
before run-time). Such a „not-predefined„ result is not necessarily a macro level
result: see for instance SWARM-like simulations. Although they can produce patterns
(of action) they do not lead to the emergence of higher-level institutions that shape
and keep a society together. A similar argument holds for the reverse direction:
macro-level structures can be implemented in a simulation statically without the need
to let them emerge.
3) Value aggregation is an analysis of macro phenomena
One way to distinguish attributes for modelling and reasoning, is to differentiate
between dimensional (i.e. numerical attributes) and structural (e.g. relationships on
cause-effect, or acquaintance, trust, influence etc.). In this differentiation the
sociological approach on the macro level (namely to look at structures) is extremely
opposed to the one used in current DAI research. The macro perspective here means
to aggregate values from the individual to the group layer and focus on dimensional
parameters like score, speed, number of communication acts, voting results etc.,
where aggregation is straightforward. The structural interpretation that could lead to
more sophisticated social reasoning, like it is done by Sichmann et al. , is rarely
4) Populations of artificial agents are artificial societies
Especially for applied multi-agent systems (the representational approach) it holds
that these agents are created with the intention of delegating actions (and in fact
delegation is viewed as a central notion in DAI: e.g. by Castelfranchi and Falcone
). In this sense many assumptions about human behaviour and the user's goals and
desires are represented by the agent acting in the multi-agent system. Therefore
observed phenomena in this population will not only be caused by artificial actors, but
also by the intentions of the human user. As a consequence it would not be correct to
speak of an artificial society, the nature of the intersection of intentions requires this
to be termed a hybrid society. In addition, sociologists would require that this
population exhibits macro aspects of the human society (see above) before it can be
considered an artificial or hybrid society.
7 Towards a Micro-Macro Definition for DAI
We are not in the position to give a final definition for the macro concept for agent-
based simulation (either foundational or representational) or decide whether the more
complex macro notion of sociology should be applied in DAI. From our research
however, we conclude that we can identify three different strands of research where
the question of the micro-macro link arises with different magnitude:
a) For moderately sized multi-agent systems (which is still the large majority of
today's applications) the list of problem definitions mentioned in Section 4 is
sufficiently complex and is most useful due to its well understood distinction in
subproblems and its precision.
b) This view is not sufficient when complexity is increased: Open and large multi-
agent systems require transfer from the social sciences in order to build systems
that are adaptive, scalable and laid out with the potential to resolve unpredictable
conflicts. A stronger notion of the macro aspects (institutions, power, fraud etc.)
becomes necessary and sociology is a source of inspiration for flexible
architectures for scalable MAS. In close analogy to the progress which AI
research has made by approaching cognitive psychology, DAI can be expected to
be brought forward by the cooperation with sociology.
c) For the knowledge transfer from DAI to the social sciences an adequate
conceptualisation of the macro aspect as it is perceived in sociology is necessary
to guide agent-based simulation and make the results transferable to sociology.
The approaches in paragraphs a) and b) can be considered representational
approaches, whereas c) corresponds to the foundational approach. Paragraph b) views
the agent as depending on features like flexibility, autonomy and social competence
(where sociologists would argue that the social ability already assumes the
Having established that for a number of problems the adoption of a complex and
well-founded notion of the macro level is desirable, we would like to discuss some
implications for future work.
A general observation from what has been said so far, is that it may be advisable to
use sociologically founded concepts, but computation of bi-directionally interacting
micro and macro-level simulation appears to be too complex and too hard to achieve
and is therefore hardly existing. When looking at this shortcoming of up-to-date social
simulation, it appears that there is a need to investigate, which sociological theory can
on the one hand improve the simulated model (e.g. the bi-directional interaction of
micro and macro) and on the other hand simplify the design of agents (frameworks
for socially more competent agents).
These are the requirements of a hybrid theory that has the explanatory power which
stretches from individual behaviour to structures of the social context and back to the
individual action. A theory that might come to mind is the theory of Anthony
Giddens. The strength of this theory lies in the concept of duality of structure and
action. Conte and Castelfranchi  criticise that although Giddens' theory „is
process-oriented, it actually does not take into sufficient account the role of the
cognitive processes linking the micro and macro levels”. In our ongoing research we
have found that the habitus-field theory of Pierre Bourdieu is a theory which covers a
similar spectrum between action and structure, while at the same time having a greater
explanatory power on the very subject that Conte and Castelfranchi describe as the
shortcoming of Giddens' theory. Bourdieu's concept of habitus consists of a set of
dispositions to actions and ways of perception. These dispositions depend on the
history of the individual and what it experienced in the past, they may be incorporated
or imitated, i.e. learned by observation and acquired by advice. We suggest that the
concept of these dispostions is a perfect starting point to connect bounded rationality
research with the DAI research of social contexts. Furthermore, the fact that Bourdieu
emphasises the practical application of his theory and has reported extensively on his
practical work, gives us the hope that his methodology can be used for application in
For Bourdieu, the habitus is the result of processes that adapt to the surrounding
social structure according to the logic of this social context. This marks the
importance and the influence of the structure of the agent society on the behaviour of
the individual, while still explaining how the individual shapes the structure. Bourdieu
views the individual with its desires and actions as the force behind the development,
change and reproduction of social structure. For us, this results in a call for more
effort in additional reasoning about structures instead of reasoning about aggregated
values for agents in social simulation. We believe that with the habitus-field theory
we have found a sociological theory that provides what Conte and Castelfranchi 
demand, when they write:
We believe that the micro-macro link is not only a two-fold issue: it is not
only a matter of relating macro-structures and micro-interactions, society and
action, as many social scientists including Giddens, seem to think. In our
conception, it is a three-faceted issue, including (a) external forces and
structures, (b) agents' cognition, and (c) their actions. Cognition plays a
fundamental linking role between the external forces and the agent's
a) unlike what is commonly called rationality, cognition reflects and
embodies in various ways objective pre-conditions, societal prescriptions and
institutions, and reinforcing effects. Cognition is undoubtedly structured by
society. The question is how is this possible?
b) macro-social phenomena may emerge, unintentionally, from micro-
interactions. However, they not only directly emerge from behaviours, but also
derive from the agent's cognitive representations and state. For example, while
some conventions directly emerge from behaviours, some structures of
interdependencies emerge from the interrelationships among the internal
properties of agents situated in a common world.
Bourdieu describes his habitus as the structure that is structured by the individuals
social context and that is also structuring the social context by the individuals
participation in this context (the „structured and structuring structure”). The
incorporation of this structure is the process of learning heuristics for action and
perception that are adequate for different contexts. According to Bourdieu these
heuristics will not be actively reconsidered before the habitus leads to a crisis. This is
an interesting pointer to learning algorithms like reinforcement learning and will
guide our future research.
We would like to thank the anonymous reviewers for their helpful criticism and ideas
on how to improve this document. We would also like to thank the participants of the
MABS workshop at the ICMAS 2000 conference for most fruitful discussions. Last
but not least, we would like to thank sociologists Michael Florian, Andrea Dederichs
and Frank Hillebrandt for their elaborate and constructive comments on an early draft.
1. Alexander, J. (1983). Theoretical Logic in Sociology. Vol. 4:Tthe Modern Reconstruction
of Classical Thought. Berkeley/Los Angeles: University of California Press.
2. Alexander, J. (1995). Fin-de-siècle Social Theory: Relativism, Reduction and the Problem
of Reason, London.
3. Bazzan, A. L. C., Bordini, R. H. und Campbell, J. A. (1997). Agents with Moral
Sentiments in an Iterated Prisoner’s Dilemma Exercise. In .
4. Berger, P. L. and Luckmann, T. (1966). The Social Construction of Reality: A Treatise in
the Sociology of Knowledge. New York, Doubleday.
5. Blumer, H. (1969). Symbolic Interactionism. Englewood Cliffs/New Jersey.
6. Bond, A. H. and Gasser, L. (1988). Readings in Distributed Artificial Intelligence, Morgan
Kaufmann, pp. 3-35.
7. Bourdieu, P. (1987). La Distinction: critique sociale du jugement. Translated by Nice, R.:
Distinction : A Social Critique of the Judgement of Taste. Harvard Univ Press. 1987
8. Castelfranchi, C. and Falcone, R. (1998). Principles of Trust for MAS: Cognitive
Anatomy, Social Importance, and Quantification. In Y. Demazeau. Proceedings of the
Third International Conference on Multi-Agent Systems (ICMAS 98), 1998.
9. Colemann, J. (1990): Foundations of Social Theory, Cambridge/Mass.
10. Colemann, J., and Bourdieu, P. (1991). Theory for changing Society, N.Y.
11. Conte, R. and Castelfranchi, C. (1996). Simulating Multi-Agent Interdependencies. A Two-
Way Approach to the Micro-Macro Link. In Troitzsch (1996).
12. Conte, R. and Moss, S. (In preparation). Agent Based Social Simulation: Technological
Roadmap. Document for the AgentLink SIG on Agent-Based Social Simulation.
13. Dautenhahn, K., Masthoff, J. und Numaoka, C. (1997). Socially Intelligent Agents. Papers
from the 1997 AAAI Fall Symposium, November 8-10, Cambridge, Massachusetts,
Technical Report FS-97-02.
14. Doran, J. E. (1997). From Computer Simulation to Artificial Societies. Transactions SCS,
vol.14(2), pp. 69-77.
15. Edmonds, B. and Dautenhahn, K. (1998). The Contribution of Society to the Construction
of Individual Intelligence. In Edmonds, B. and Dautenhahn, K. (eds.), Socially Situated
Intelligence: a workshop held at SAB'98, August 1998, Zürich. University of Zürich
Technical Report, 42-60.
16. Giddens, A. (1984). The constitution of Society, Cambridge.
17. Goffmann, E. (1967). Interaction Ritual. New York: Anchor Books.
18. Goffmann, E. (1959). The presentation of Self in Everyday-Life, N.Y.
19. Goffmann, E. (1961). Encounters, Indianapolis, N.Y.
20. Hannoun, M., Sichman, J., Boissier, O. and Sayettat, C. (1998). Dependence Relations
Between roles in a Multi-Agent System: Towards the Detection of Inconsistencies in
Organization. In Sichman, J., Conte, R. and Gilbert, N. (eds.), Multi-Agent Systems and
Agent-Based Simulation, LNAI 1534.
21. Homans, C. (1950). The human group, New York.
22. Homans, C. (1974). Social Behavior, New York.
23. Kennedy, J. (1999). Artificial Sociocognition: Cultures of Belief in Populations of
Elementals. In Landauer, C. and Bellman, K. L. (eds.): Virtual Worlds and Simulation
Conference (VWSIM'99), Simulation Series, vol 31 (2), The Society for Computer
24. Jennings, N. R. and Campos, J.R. (1997). Towards a social level characterisation of
socially responsibel agents", IEE proceedings on software engineering 144(1), pp. 11-25.
25. Jennings, N. R., Sycara, K. and Wooldridge, M. J. (1998). A roadmap of agent research
and development. Autonomous Agents and Multi-Agent Systems, 1:7-38.
26. Kuhn, T. (1970). The Structure of Scientific Revolutions, Chicago.
27. Langton, C. B. (1989). Artificial Life. Studies in the Sciences of Compexity, vol. 6, pp. 1-
28. Luhmann, N. (1995). Social Systems. Stanford University Press.
29. Luhmann, N. (1997). Die Gesellschaft der Gesellschaft. Suhrkamp.
30. Marcenac, P., Courier, R., Calderoni, S. and Soulie, C. (1998). Towards an Emergence
Machine for Complex Systems Simulations. In Del Pbil, A. P. and Mira, J. and Ali, M.
(eds.): Lecture Notes in Artificial Intelligence, vol. 1416, Springer Verlag.
31. Malsch, T. (2000). From a talk given at the kick-off meeting of the German socionics
projects in Hamburg Rissen, January 28th.
32. Mead, G. (1934). Mind, Self and Society, Chicago.
33. Moulin, B. and Chaib-Draa, B. (1996). An overview of distributed artificial intelligence.
In O'Hare, G. M. P. and Jennings, N. R. (eds.): Foundations of Distributed Artificial
Intelligence, pp. 3-55. John Wiley & Sons Inc., New York.
34. Müller, H.J., Malsch, T. and Schulz-Schaefer, I. (1998). SOCIONICS: Introduction and
Potential. Journal of Artificial Societies and Social Simulation. vol. 1(3).
35. Nwana, H. S. (1996). Software Agents: An Overview. In Knowledge Engineering Review,
vol. 11 (3), pp. 205-244.
36. Parsons, T. and Shills, C. (1962). Toward a General Theory of Action. New York.
37. Parsons, T. (1951). The social system. Routledge & Kegan Paul, London.
38. Parsons, T. (1971). The system of modern societies. Englewood Cliffs, N.J.
39. Sandholm, T. (1999). Distributed Rational Decision Making. In Weiss (1999).
40. Sichman, J.S., Conte, R., Castelfranchi, C., and Demazeau, Y. (1994). A social reasoning
mechanism based on dependence networks. Proceedings of the 11th European Conference
on Artificial Intelligence, Amsterdam.
41. Schillo, M., Rovatsos, M. and Funk, P. (2000). Using trust for Detecting Deceitful Agents
in Artificial Societies. In the AAI Journal, Special Issue on "Deception, Fraud and Trust in
Agent Societies”, eds: Castelfranchi, C., Tan, Y., Falcone, R., and Firozabadi, B. S.
42. Schütz, A. and Luckmann, T. (1973). The Structures of Life-World. London: Northwestern
43. Troitzsch, K., Mueller, U., Gilbert, N. and Doran, J. (1996). Social Science
44. Troitzsch, K. (1996). Multilevel Simulation. In .
45. Verhagen, H. and Smit, R. (1997). Multi-agent systems as simulation tools for social
theory testing. Poster presentation at International Conference on Computer Simulation
and the Social Sciences (ISSC&SS), Cortona.
46. Wallerstein, I. (1991). The Modern World-System: Capitalistic Agriculture and the
Origins of the European World-Economy in the sixteenth Century. Academic Press.
47. Weiß, G. (1999). Multi-Agent Systems: A Modern Approach to Distributed Artificial
Intelligence. MIT Press.
48. Yamagishi, T. (1986) The Provision of a Sanctioning System as a Public Good. In Journal
of Personality and Social Psychology, vol. 51 (1), pp. 110-116.