Contract Enactment in Virtual Organizations: A Commitment-Based Approach∗
Yathiraj B. Udupi and Munindar P. Singh
Department of Computer Science
North Carolina State University
Raleigh, NC 27695-8206,USA
A virtual organization(VO) isadynamic collection of entities
(individuals, enterprises, and information resources) collabo-
rating on some computational activity. VOs are an emerging
means tomodel, enact, and manage large-scalecomputations.
VOs consist of autonomous, heterogeneous members, often
dynamic exhibiting complex behaviors. Thus, VOs are best
modeled via multiagent systems. An agent can be an indi-
vidual such as a person, business partner, or a resource. An
agent may also be a VO. A VO is an agent that comprises
Contracts provide a natural arms-length abstraction for mod-
eling interaction among autonomous and heterogeneous
agents. The interplay of contracts and VOs is the subject of
this paper. The core of this paper is an approach to formalize
VOs and contracts based on commitments.
Our main contributions are (1) a formalization of VOs, (2) a
discussion of certain key properties of VOs, and (3) an iden-
tification of a variety of VO structures and an analysis of how
they support contract enactment. We evaluate our approach
with an analysis of several scenarios involving the handling
of exceptions and conflicts in contracts.
Virtual organizations (VOs) are dynamic collaborative col-
lections of individuals, enterprises, and information re-
sources (Foster, Kesselman, & Tuecke 2001). Traditionally
such collaborative activities are focused on data sharing and
computation. This paper emphasizes VOs in business set-
tings, especially where processes support delivery of real-
world (not just IT) services. Production grids employed for
scientific or business computing are excellent examples of
settings where this approach can be applied. Because of le-
gal and economic pressures, business environments provide
richer policies than the more common scientific computing
environments. VOs, whether business or scientific, have key
propertiesthat distinguish them from traditionalIT architec-
Autonomy. The members of a VO behave independently,
constrained only by their contracts.
∗This research was supported by the National Science Founda-
tion under grant ITR-0081742.
Copyright c ? 2006, American Association for Artificial Intelli-
gence (www.aaai.org). All rights reserved.
Heterogeneity. The members of a VO are independently
designed and constructed, constrained only by the appli-
cable interface descriptions.
Dynamism. The configuration of a VO changes at runtime
as members join and leave.
Structure. VOs have complex internal structures, reflected
in the relationships among their members.
The above properties of VOs closely match the properties of
multiagentsystems. Agentsare persistent computationsrep-
resenting independent principals: they are autonomous and
heterogeneous as a result. Multiagent systems are motivated
from flexible human organizations and consequently exhibit
dynamism and structure. Thus the distinguishing properties
of VOs are mirrored in multiagent systems.
Collaborations among agents are structured via contracts.
A contract is modeled as a set of commitments. A VO is
formed between the contracting agents if it does not exist
already. VOs can have complex nested structures and hence
contracts may be formed at multiple levels. More than one
contract may simultaneously exist among a set of contract-
ing agents. Here, the VOs within which the contracts are
formed may overlap resulting in situations where an agent
belongs to two or more VOs, neither of which is an an-
cestor of the other. Several other factors come into play
while creating contracts in addition to the consequences of
the VO structures. VOs have key properties that are essen-
tial for handling contracts, commitments, and the various
operations on commitments. This paper identifies several
different VO structures and their implications on contract
enactment and vice versa.
A VO not only encapsulates some relationships among its
members, but also functions as an agent that engages in po-
tentially complex relationships with its members. In our ap-
proach, these relationships are expressed in terms of goals,
policies, and commitments. For example, the goals of a VO
can be propagated to its members as goals, or may become
the commitments of its members. Likewise the policies of a
VOwould normallybepropagatedto its members. Thepoli-
cies of a VO might control how the commitments among its
members evolve. Consequently, as an important example, if
two agents enter into a contract, besides the commitments
that are explicitly part of the contract, their behavior would
be constrained by the goals, policies, and commitments of
mental attitudes (beliefs, desires, and goals) of an organiza-
tion. Our approach formalizes contracts in terms of com-
mitments, and demonstrates the enactment of the contracts
based on operations on commitments. The commitments in
our approach are similar in some respects to Boella et al.’s
notions of obligations. However, our approach relates con-
tracts to commitmentoperations,andthus yields a morepre-
cise model of the enactment of contracts.
are patterns of organizing multi-
agent system with a view to classifying their performance
characteristics (Horling & Lesser 2005). Horling et al.
present a distributed algorithm that uses an underlying or-
ganization to guide coalition formation. Brooks and Durfee
(2003) demonstrate how “congregations”of agents can ben-
efit multiagent systems in searching other agents and mini-
mizing the interactions and search costs.
are formed when a number of cooperative agents
get together to accomplish a common goal (Tambe 2003).
Agents coordinate their actions in a way that is consistent
and supportive of their team’s goal. Our approach consid-
ers commitment-based contracts that bring together agents
to form VOs. Here, the contracting agents collaborate in
the context of a common VO to accomplish the contract
goals. Our approach can be thought of as addressing the
same basic problem, robust teamwork, but specialized to
VOs where contracts capture the essence of the interactions
andthe commitmentoperationssupportresponsesto various
Social reasoning mechanisms and relationships enable an
agent to evaluate and reason about others using its depen-
dencies with others (Sichman & Conte 2002). Several rela-
tionships can exist among the agents and these relationships
caninfluencethe actionstakenby them. This ideacan been-
hanced and combined with our approach as follows. In the
VO context, relationships can exist between a VO and its
children, as well as among the siblings. Relationships here
are dynamic, because they can be formed and revoked at run
time. Such relationships can form the basis of policies, and
can influence the operations on commitments and contracts.
This paper proposes a commitment-based architecture for
VOs. This architecture treats VOs as consisting of agents,
potentially VOs in their own right. The nesting structure of
the VOs highlights the freedoms and constraints on the VOs
at each level. Each VO is associated with a set of goals,
commitments, and policies. The key advantages of this ar-
chitecture are as follows.
Relationships. The proposed architecture naturally sup-
ports complex structures. It enables VOs that are nested
or partially overlapping. It supports managing the com-
plementary properties of two VOs being unaware of each
other’s structure but gaining requisite visibility to interact
Policy Management. The proposed architecture recog-
nizes that VOs are distributed. It supports two comple-
mentary perspectives. One is that there is a single locus
of policy enforcement. The other is that a distributed or-
ganization must have parts that collaborate to enforce a
Contract Enactment. The proposed architecture provides
a commitment-based argument for contract enactment.
Our hierarchical and distributed VO setup facilitates han-
dling of various scenarios of conflicts and exceptions in
The effect of relationships between agents on contract en-
actment in a VO and other enhancements to our formal VO
definitions will be considered as near future work. The re-
lationships captured while describing VOs can dynamically
change and becomes crucial for describing VO behaviors.
Boella, G., and van der Torre, L. 2004. Contracts as le-
gal institutions in organizations of autonomous agents. In
Proceedings of the 3rd International Joint Conference on
Autonomous Agents and Multiagent Systems, 948–955.
Boella, G.; Hulstijn, J.; and van der Torre, L. 2005. Vir-
tualorganizationsas normativemultiagentsystems. InPro-
ceedings of the 38th Annual Hawaii International Confer-
ence on System Sciences, 192–201.
Brooks, C. H., and Durfee, E. H. 2003. Congregation
Formationin MultiagentSystems. AutonomousAgentsand
Multi-Agent Systems 7(1-2):145–170.
Dastani, M.; Dignum, V.; and Dignum, F. 2002. Organi-
zations and normative agents. In Proceedings of the First
EurAsian Conference on Information and Communication
Technology, 982–989. Springer-Verlag.
Feeney, K. C.; Lewis, D.; and Wade, V. P. 2004. Policy
based management for Internet communities. In Proceed-
ings of 5th International IEEE Workshop on Policies for
Distributed Systems and Network (POLICY), 23–32.
Foster, I.; Kesselman, C.; and Tuecke, S.
anatomy of the grid: Enabling scalable virtual organiza-
The International Journal of High Performance
Computing Applications 15(3):200–222.
Horling, B., and Lesser, V. 2005. A Surveyof Multi-Agent
Organizational Paradigms. The Knowledge Engineering
Sichman, J. S., and Conte, R. 2002. Multi-agent depen-
dence by dependence graphs. In Proceedings of 1st In-
ternational Joint Conference on Autonomous Agents and
Multiagent Systems, 483–490.
Singh, M. P. 1999. An ontologyfor commitments in multi-
agentsystems: Towarda unificationofnormativeconcepts.
Artificial Intelligence and Law 7:97–113.
Tambe, M. 2003. Towards flexible teamwork. Journal of
Artificial Intelligence Research 7:83–124.