Modelling Multi-Agent Systems

with Organizations in Mind

Matthias Wester-Ebbinghaus and Daniel Moldt

University of Hamburg, Department of Informatics

Vogt-K¨olln-Straße 30, D-22527 Hamburg, Germany

Abstract. Software systems are subject to increasing complexity and in need of

efﬁcient structuring. Multi-agent system research has come up with approaches

for an organization-oriented comprehension of software systems. However, when

it comes to the collective level of organizational analysis, multi-agent system

technology lacks clear development concepts. To overcome this problem while

preserving the earnings of the agent-oriented approach, this paper propagates

a shift in perspective from the individual agent to the organization as the core

metaphor of software engineering targeting at very large systems. According to

different levels of analysis drawn from organization theory, different types of or-

ganizational units are incorporated into a reference architecture for organization-

oriented software systems.

1 Introduction

Modern software systems are subject to ever increasing size and complexity. Within the

IT community the expectation begins to form that the sheer size and speed of growth

of these systems render most traditional software engineering approaches (relying on

a top-down design, expecting comprehensive knowledge of relevant system-wide pa-

rameters, based on the possibility of applying a central control facility) inept [16]. It is

these software systems in the large that the article at hand is devoted to.

Multi-agent systems as a software engineering paradigm are one candidate to pro-

vide solutions for this kind of systems. Exemplary in [11], Jennings argues that multi-

agent systems are very well suited for the realization of three particularly important

techniques for handling complex software, namely decomposition, abstraction, and or-

ganization. At the same time however, Jennings calls for a social level view on multi-

agent systems in order to deal with the difﬁculties that agent autonomy and sophistica-

tion in combination pose on the prediction of the overall system behaviour by leading

to a considerable scope of emergence.

In subsequent years until today various approaches have been brought forth that are

in line with this request by taking the perspective on a multi-agent system as an organi-

zation (an overview of recent and current work can be found in [19]). Noteworthy, the

rationale for adopting an organization-oriented perspective throughout the approaches

mostly refer to the very same features that Hannan and Carroll identify as the main

capacities of organizations in human societies: Organizations are durable, reliable and

Wester-Ebbinghaus M. and Moldt D. (2008).

Modelling Multi-Agent Systems with Organizations in Mind.

In Proceedings of the 6th International Workshop on Modelling, Simulation, Veriﬁcation and Validation of Enterprise Information Systems, pages 81-90

DOI: 10.5220/0001737900810090

 SciTePress

accountable [10].

In this respect, organization-oriented approaches to multi-agent sys-

tem engineering seek to combine local agent autonomy with the assurance of global

system characteristics by imposing ”organizational facts” onto the system. Boissier [3]

identiﬁes different organizational dimensions (with structural, functional and interac-

tional dimensions as the most prominent ones) that receive varying emphasis depending

on the particular approach.

However, when relating multi-agent system approaches to organization theory it

becomes obvious that the true potential of the organizational metaphor is not entirely

exploited. Multi-agent system research so far has mainly focussed on the conception

of organizations as contexts for individual agents. As we have analyzed in [19], the

importance of organizations as corporate actors that Scott [18] stresses for more global

(ecological) levels of analysis has been largely neglected.

Consequently, the long term goal of our work is the provision of a software de-

velopment approach that builds upon and extends the multi-agent system approach in

order to account for the true potential of the organizational metaphor. In this paper we

supplement this goal by proposing an abstract reference architecture for organization-

oriented software systems. Ferber [8] advances the distinction between ACMAS (agent-

centred multi-agent systems) and OCMAS (organization-centred multi-agent systems).

We consider our approach as one further step in this shift of paradigm from agent- to

organization-orientationand term the systems introduced by our approach MOS (multi-

organization systems).

In Section 2 we present our approach of modelling open and controlled system

units. We utilize the introduced universal scheme to propose a reference architecture

for multi-organization systems composed of concrete organizational units in Section 3.

We conclude our results and provide an outlook to future work in Section 4.

2 Open System Modelling

The introduction should have made clear that the software systems addressed here are to

be comprehended as systems of systems. To arrive at an illustrative modelling approach

despite the inherent complexity of the systems in focus we adopt the modular view of

each system as a unit that maintains relationships to other system units. As a prelimi-

nary step to dealing with particular types of system units in our organization-oriented

architecture, this section refers to system theory in order to introduce a general model.

2.1 Basic System Unit

Most of the important entities studied by scientists – nuclear particles, atoms, molecules,

cells, organs, organisms, communities, organizations, societies, solar systems – come

under the category of a system [2]. Consequently, the characterization of a system fol-

lows quite abstract as an assemblage or combination of parts whose relations make

them interdependent.

It will not be discussed in this paper that each of these capacities is a double-edged sword and

not all organizations measure up to them.

add remove

use / modify

internal system parts

Fig.1. An Abstract View on System Units.

Based on this abstract characterization, Figure 1 shows a general model of system

units.

The evolution of the overall system unit solely depends on its internal parts, whether

new parts are added (add), former parts are removed (remove) or current parts are used

and potentially modiﬁed (modify / use). The details of these operations and how they

come into being depend on the characteristics of a particular system.

2.2 Recursive Nesting

The basic model of system units of the former subsection emphasizes the similarities of

all types of systems. However, there exist of course substantial differences between par-

ticular systems. Exemplary, Boulding [4] presents a typology of systems that advances

from physical over biological to social systems. Along the way, each successive system

becomes progressively more complex, more loosely coupled, more dependent on infor-

mation ﬂows, more capable of self-maintenance and renewal, more able to grow and

change, and more open to the environment.

Especially the openness to the environment is of particular importance regarding

the software systems this paper addresses. Systems of systems implicate a network of

relations where each system (to different degrees) relies on the services and resources of

other systems. Interactions with the environment and throughput of external resources

are regarded as crucial for the functioning and self-maintenance of open systems [17,

5]. Nevertheless, open systems have boundaries and spend energies to maintain them.

Consequently, one can identify the twin properties of open systems as consisting of two

basic (and opposing) sets of system processes [5]. The term morphostasis refers to those

processes that tend to preserve or maintain a system’s givenform, structure, or state. The

term morphogenesis refers to processes that elaborate or change the system. While both

are not exclusive to open systems they receive a special emphasis. In adapting to the

external environment, open systems typically become more differentiated in form and

more elaborate in structure.

It is not very helpful to regard the environment of an open system as simply ”ev-

erything else”. It suggests itself to comprehend the environment again as a system (or

multiple systems) and thus to arrive at a recursive understanding of open system models

as for example explicated by Koestler’s concept of a holon [12] and in Beer’s recursive

model of viable systems [1]. Each open system is characterized as ”Janus-faced” with

The model has a coloured Petri nets semantics, cf. [9].

add

remove

modify

integration

operation

governance

Fig.2. System Unit with Coloured Internal Parts.

an inner eye at the internal systems and an outer eye at the surrounding system (or sys-

tems). Here, our perspective on systems as units leads to an illustrative understanding.

The relation of open systems to their environments is traced back to the (not necessarily

unique or disjoint) nesting of system units, the embedding of system units inside other

system units.

We use this short summary of system theoretical inspirations to reﬁne our model

of system units. The ﬁrst step is a colouring of the internal parts of a system unit as

shown in Figure 2.

The internal parts are now explicitly to be regarded as system units

themselves. The colouring is no partitioning, the different sets of internal units need not

be disjoint. It is more of a conceptual classiﬁcation according to function.

The operational units are those parts of the system in focus that undertake the sys-

tem’s primary activities. In a manufacturing setting they would be the production units,

teams of people, and machines that actually do the manufacturing. In a complex pro-

duction organization they would include manufacturing, distribution, and warehousing.

Basically, the operational units are the intrinsic parts of the system.

The integrational units see to it that the singular operational units are integrated into

a joint system in the ﬁrst place. They deﬁne the means by which the operational units

may participate in the system and regulate their activities. One example is the nervous

system of the human organism that connects the muscles and organs (being operational

units). In another setting, they embody hierarchical planning and performance control

systems of an enterprise.

Governance units are responsible for certain system processes and structures being

in place, to ensure the adherence to system laws, and to maintain mechanisms for con-

trol and coercion. One example is the brain of the human organism that oversees the

complex of muscles and organs and tries to optimze them. It also establishes a connec-

tion to the environment through its senses. It plans, projects and develops an identity.

The transitions connecting to the outer circle are just short forms that include all three cases

for the inner circles respectively. To obtain a well-formed Petri net model the short forms have

to be resolved, resulting in a total of nine transitions for Figure 2.

Another example is the board of directors of an enterprise that sets the goals and strate-

gies, determines the budget and establishes connections to core business partners.

A clear separation between the different types of internal system units is not always

possible just as it is not always clear where to locate morphostasis and morphogenesis.

For example,in one system the purposeof the governanceunits might be to just preserve

and maintain a set of largely ﬁxed system laws. In another case, the government units

might be a continuous source of major renewal.

Generally speaking, the distinction serves to carry out separation of concerns in two

ways. Firstly, the administrative units are distinguished from the operational units that

they embed. Secondly, technical embedding (via the integrational units) is distinguished

from strategical embedding (via the governance units).

2.3 Structure in Threes

The overall behaviour of a system unit manifests in various system processes that shape

and evolve the system. In order to study these processes exhaustively, we identify no

less than 27 case distinctions that stem from three orthogonal dimensions with three

values respectively.

– Operation: The three basic operations that a system process may relate to are add,

remove, and use/modify.

– Direction: System processes may impact the system from three conceptual direc-

tions. The operational units (enabled by the integrational units) impact the system

from below. The maintenance units impact the system at the same level. Finally,

surrounding system units may impact the system in focus from above.

– Affected internal system unit: Each system process may involve all three kinds of

system units, operation, integration, and governance.

The details of each case and whether it is of relevance at all depend on a particu-

lar system. Consequently, there is no point in addressing each individual case for our

general model of system units. Instead, we provide coarsened models for the three di-

rections that summarize multiple cases of the other two dimensions.

Figures 3 - 5 show the inﬂuences on the system from below, at the same level,

and from above through the integration of operational units, controlling activities by

governance units and peripheral connections to surrounding system units respectively.

To obtain a modelling for each of the 27 cases the ﬁrst step is to reﬁne the transitions

that connect to the outer circle of the system unit in focus. In addition, the three cases

shown here are somewhat ”pure” cases. In particular systems they are typically merged.

3 Reference Architecture

The reﬁned model of open system units still does not offer a meaningful architectural

model for IT systems. It lacks differentiation for particular perspectives. The central

concern of this section (and the overall paper at hand) is to advance a proposal for

software architectures in the large based on the universal scheme of open system units

and their embedding inside each other.

offer frame

integration

operation

governance

arise

vanish

Fig.3. Open System Unit: Integration Processes.

create

remove

govern

integration

operation

governance

Fig.4. Open System Unit: Governance Processes.

The modelling of complex systems requires different levels of abstraction. The con-

tents at each level should be described in a way that offers a largely complete and

homogenous picture of the selected perspective for this level of abstraction. From this

premise we derive our architectural proposal for multi-organization systems.

3.1 Overview

In focussing on the organization as the core unit of the architecture, three necessary

levels of examination directly follow: The organization itself, its internals and its envi-

ronment. The architecture shall include multiple organizations and each of these may

have different (conceptual) environments. It follows the necessity of a fourth architec-

tural level as a system closure for the integration of all environments.

Interestingly, this identiﬁcation of four levels resulting from rather technical consid-

erations is conﬁrmed by organization theory according to Scott [18]. The internals of

an organization correspond to the socio-psychological level of organization theoretical

analysis where the behaviour and relations between individual members of the organi-

zation is examined. From this perspective, organizations are regarded as contexts. The

organization as a discrete entity of its own appears at the organization structure level

enter

leave

use frame

integration

operation

governance

Fig.5. Open System Unit: Peripheral Processes.

where the structural properties and social processes that characterize an organization

and its subdivisions are studied. The ecological level ﬁnally focuses on characteristics

and actions of organizations as corporate actors that operate in even more global net-

works of relations. For the ecological level, further reﬁnements are possible. Among

these the concept of organizational ﬁeld is the most comprehensive one referring to

immediate environments of organizations while the society offers a common frame for

organizational ﬁelds.

As a consequence, our reference architecture for multi-organization systems con-

sists of four levels of system units. In order to emphasize that these units are particu-

lar instances of the universal scheme of system units according to Section 2, they are

termed organizational units. Figure 6 shows an overview of the architecture as nested

organizational units.

Each department is exclusively assigned to an organization. Each organization

consists of multiple departments and operates on multiple organizational ﬁelds. Each

ﬁeld hosts multiple organizations and is embedded in the single integrative society.

Departments as the lowest level units of abstraction embody the connection to multi-

agent system technology. Each department is a multi-agent system. This perspective

only makes sense if a department fulﬁls an organizational purpose. As described in the

introduction of this paper, a considerable part of multi-agent system research of the past

years was devoted exactly to this aspect. Thus, we arrive at a seamless transition from

agent- to organization-orientation. All organizational units of the reference architecture

can be regarded as logical units (nevertheless embodied by explicit software constructs,

e.g. being agentiﬁed) that are built upon physical multi-agent systems.

This distinction according to Scott is a reﬁnement of the prevailing distinction in micro- and

macro-level where the ﬁrst corresponds to the social-psychological level and the latter encom-

passes all other levels.

The model has a reference net semantics. Reference nets show some extensions compared to

conventional coloured Petri nets [14]. They implement the nets-within-nets paradigm where a

surrounding net (the system net) can have nets as tokens (the object nets). Reference semantics

is applied, so these tokens are references to net instances. Synchronous channels allow for

communication between net instances.

government

infrastructure

fields

institution

material-ressource

structure

inhabitants

dominant

coalition

superstructure

subdivisions

management

positions +

grouping

members

organizational

field

department

society

organization

ZOOM

Fig.6. Multi-Organization Software Architecture: Overview.

3.2 Architectural Levels

Due to space limitations, no complete description of the architectural levels is possible

in this paper. Instead, we present a short summary of the characteristics of each level in

order to sketch the conceptual differences.

– Society. The society as the highest architectural level embodies the closure of the

system and thus has no super-ordinate units. It embeds organizational ﬁelds. These

are connected through a ﬁeld infrastructure for interaction and migration between

ﬁelds. The government is a legal authority that holds and enforces system-wide

(ﬁeld-spanning) societal laws.

– OrganizationalField. DiMaggio characterizes organizationalﬁelds as critical units

bridging the organization and society level in the study of social and community

change [7]. For example, state regulations directed at individual organizations are

typically mediated by ﬁeld level structures such as trade associations.

Concerning the characteristics of an organizational ﬁeld, two broad categories are

distinguished. The material-resource structure characterizes the ﬁeld as a stock

of resources and source of information, which provide the primary premise under

which organizations as inhabitants of the ﬁeld come together. However, material-

resource environmentsalways rest on an institution. As Scott puts it, institutions are

composed of regulative, normative, and cultural-cognitive elements that together

with associated activities and resources provide stability to social life [18].

– Organization. Organizations put particular emphasis on an organizationalstructure

and an organizational authority. Mintzberg [15] for example identiﬁes ﬁve funda-

mental types of organizational subdivisions (operating core, middle line, strategic

apex, technostructure, support staff) that are integrated into an organizational su-

perstructure. It is built up by grouping individual positions into units and units

into ever larger units until the hierarchy is complete.

Each organization has an authority that is in charge of power and setting the or-

ganizational goals. Cyert and March [6] come up with a quite broad concept. Or-

ganizations are viewed as being composed of various and varying coalitions, each

of which seeks to impose its preferences onto the larger system. If none of them

succeeds, they seek as allies other coalitions whose interests are related. Finally, a

conglomerate will arrive at a mutually acceptable agreement and at the same time

will be inﬂuential enough to constitute the dominant coalition of the organization.

– Department. The requirement of departments being exclusively assigned to orga-

nizations is a logical one. Departments of different organizations need not be dis-

joint and might for example acquire their members from the same physical multi-

agent system. Nonetheless, it is crucial to distinguish between different depart-

ments. This issue has been addressed by multi-agent system technology and cor-

responding solutions follow the common organization implementation architecture

for open MAS from [3]. The integrational units as positions and grouping charac-

teristics represent an organizational layer that encapsulates organizational speciﬁ-

cations. This layer offers proxies to which domain agents from an open multi-agent

system must connect to act as members in the organization.

Supervision and authoritarian decision making might be woven into the position

and grouping speciﬁcations or might instead (or additionally) be taken care of by

an explicit management.

4 Conclusions

We have presented an extended perspective on current organization-oriented multi-

agent system engineering and derived a reference architecture for multi-organization

systems. The architecture introduces four types of (logical) organizational units built

upon (physical) multi-agent systems.

The rationale for the selection of the particular organizational units of the archi-

tecture is deeply rooted in organization theory. The result is a software engineering

approach that supports micro as well as macro perspectives and at the same time is ac-

companied by concepts and constructs that are familiar from real-world social scenar-

ios. In this respect, our proposal is also one step in the direction of supporting a proper

IT alignment through a homomorphism between real-world and software artefacts.

As a related aspect, we consider our approach to be multi-perspective. One partic-

ular software system might appear in multiple instances of multiple types of organiza-

tional units at the same time. It all comes down to embedding relations. For example,

one software system might relate to a second one like an organization to a ﬁeld and to

a third one like a department to an organization (virtual organization).

Turning to future work, the practical usage of the architecture is the most pressing

issue. As a starting point, a Petri net-based model of organizational structures and ser-

vices is presented in [13]. At the same time it is demonstrated how agent technology

can be used as a middleware to deploy the organizational speciﬁcations. The modelling

approach is general enough to be adapted for arbitrary levels of abstraction and allows

to deﬁne collective entities and nest them inside each other. Thus it supports the devel-

opment of multi-level architectures.

References

1. S. Beer. Brain of the Firm. John Wiley & Sons, second edition, 1994.

2. L. v. Bertalanffy. General system theory. In General Systems: Yearbook of the Society for

the Advancement of General Systems Theory, volume 1, pages 1–10. Ann Arbor, MI: The

Society, 1956.

3. O. Boissier, J. H¨ubner, and J. S. Sichman. Organization oriented programming: From closed

to open systems. In Proceedings of the Seventh International Workshop on Engineering

Societies in the Agents World (EASW 2006), 2006.

4. K. Boulding. General systems theory: The skeleton of science. Management Science, 2:197–

208, 1956.

5. W. Buckley. Sociology and Modern Systems Theory. Upper Saddle River, NJ: Prentice Hall,

1976.

6. R. Cyert and J. March. A Behavioral Theory of the Firm. River, NJ: Prentice Hall, 1963.

7. P. DiMaggio. Structural analysis of organizational ﬁelds: A blockmodel approach. In Re-

search in Organizational Behaviour, volume 8, pages 355–370. Greenwich, CT: JAI Press,

1986.

8. J. Ferber, O. Gutknecht, and F. Michel. From agents to organizations: an organizational

view of multi-agent systems. In Agent-Oriented Software Engineering IV, 4th International

Workshop, AOSE 2003, volume 2935 of Lecture Notes in Computer Science. Springer Verlag,

2003.

9. C. Girault and R. Valk. Petri nets for systems engineering: a guide to modelling, veriﬁcation

and applications. Springer Verlag, 2003.

10. M. Hannan and G. Carroll. An introduction to organizational ecology. In Organizations

in Industry: Strategy, Structure and Selection, pages 17–31. New York: Oxford University

Press, 1995.

11. N. Jennings. On agent-based software engineering. Artiﬁcial Intelligence, 177(2):277–296,

2000.

12. A. Koestler. The Ghost in the Machine. Henry Regnery Co., 1967.

13. M. K¨ohler and M. Wester-Ebbinghaus. Petri net-based speciﬁcation and deployment of orga-

nizational models. In Proceedings of the International Workshop on Petri Nets and Software

Engineering (PNSE’07), pages 67–81, Siedlce, Poland, June 2007. Akademia Podlaska.

14. O. Kummer. Referenznetze. Logos Verlag, Berlin, 2002.

15. H. Mintzberg. Structure in Fives: Designing Effective Organizations. Prentice-Hall, 1983.

16. L. Northrop. Ultra-Large-Scale Systems: The Software Challenge of the Future. Software

Engineering Institute, Carnegie Mellon, 2006.

17. L. Pondy and I. Mitroff. Beyond open system models of organization. In Research in Orga-

nizational Behaviour, volume 1, pages 3–39. CT: JAI Press, 1979.

18. W. R. Scott. Organizations: Rational, Natural and Open Systems. Prentice Hall, 2003.

19. M. Wester-Ebbinghaus, D. Moldt, C. Reese, and K. Markwardt. Towards Organization–

Oriented Software Engineering. In Software Engineering Konferenz 2007 in Hamburg:

SE’07 Proceedings, volume 105 of LNI, pages 205–217. GI, 2007.