A Model for the Structural, Functional, and Deontic Specification of Organizations in Multiagent Systems (MOISE+)

Source: Jomi Fred Hübner, Jaime Simão Sichman, Olivier Boissier; SBIA 2002, LNAI 2507 (Springer) · 2002

Abstract. A Multiagent System (MAS) that explicitly represents its organization normally focuses either on the functioning or the structure of this organization. However, addressing both aspects is a prolific approach when one wants to design or describe a MAS organization. The problem is to define these aspects in such a way that they can be both assembled in a single coherent specification. The MOISE+ model – described here through a soccer team example – intends to be a step in this direction since the organization is seen under three points of view: structural, functional, and deontic.

Introduction. The organizational specification of a Multiagent System (MAS) is useful to improve the efficiency of the system since the organization constrains the agents behaviors towards those that are socially intended: their global common purpose [8,7]. Without some degree of organization, the agents’ autonomy may lead the system to lose global congruence. The models used to describe or project an organization are classically divided in two points of view: agent centered or organization centered [10]. While the former takes the agents as the engine for the organization formation, the latter sees the opposite direction: the organization exists a priori (defined by the designer or by the agents themselves) and the agents ought to follow it. In addition to this classification, we propose to group these organizational models in (i) those that stress the society’s global plans (or tasks) [12,11,13] and (ii) those that have their focus on the society’s roles [5,6,9]. The first group concern is the functioning of the organization, for instance, the specification of global plans, policies to allocate tasks to agents, the coordination to execute a plan, and the quality (time consumption, resources usage, ... ) of a plan. In this group, the global purposes are better achieved because the MAS has a kind of organizational memory where the best plans to achieve a global goal are stored. On the other hand, the second group deals with the specification of a more static aspect of the organization: its structure, i.e., the roles, the relations among them (e.g.: communication, authority), roles obligations and permissions, group of roles, etc. In these latter models, the global purpose is accomplished while the agents have to follow the obligations and permissions their roles entitle them. Thus we should state that organization models usually take into account either the functional (the first group) or structural (second group) dimension of the organization. However, in both groups the system may or may not have an explicit description of its organization that allows the organizational centered point of view.

The Fig. 1 briefly shows how an organization could explain or constrain the agents behavior in case we consider an organization as having both structural and functional dimensions. In this figure, it is supposed that a MAS has the purpose of maintaining its behavior in the set P, where P represents all behaviors which draw the MAS’s global purposes. In the same figure, the set E represents all possible behaviors in the current environment. The organizational structure is formed, for example, by roles, groups, and links that constrain the agents behavior to those inside the set S, i.e., the set of possible behaviors (E ∩S) becomes closer to P. It is a matter of the agents, and not of the organization, to conduct their behaviors from a point in ((E ∩S) −P) to a point in P. In order to help the agents in this task, the functional dimension contains a set of global plans that has been proved efficient ways of turning the P behaviors active. For example, in a soccer team one can specify both the structure (defense group, attack group, each group with some roles) and the functioning of the team (e.g.: rehearsed plays, as a kind of predefined plans that was already validated).

If only the functional dimension is specified, the organization has nothing to “tell” to the agents when no plan can be performed (the set of possible behavior is outside the set F of the Fig. 1). Otherwise, if only the organizational structure is specified, the agents have to reason for a global plan every time they want to play together. Even with a smaller search space of possible plans, since the structure constrains the agents options, this may be a hard problem. Furthermore, the plans developed for a problem are lost, since there is no organizational memory to store these plans. Thus, in the context of some application domains, we hypothesize that if the organization model specifies both dimensions while maintaining a suitable independence among them, then the MAS that follows such a model can be more effective in leading the group behavior to its purpose (Fig. 1). Another advantage of having both specifications is that the agents can reason about the others and their organization regarding these two dimensions in order to better interact with them (in the case, for example, of social reasoning).

Related work. A first attempt to join roles with plans is the MOISE (Model of Organization for multI-agent SystEms). The MOISE is structured along three levels: (i) the behaviors that an agent is responsible for when it adopts a role (individual level), (ii) the interconnections between roles (social level), and (iii) the aggregation of roles in large structures (collective level)[9]. The main shortcoming of MOISE, which motivates its extension, is the lack of the concept of an explicit global plan in the model and the strong dependence among the structure and the functioning. This article sets out a proposal for an organizational model, called MOISE+, that considers the structure, the functioning, and the deontic relation among them to explain how a MAS organization collaborates for its purpose. The objective is an organization centered model where the first two dimensions can be specified almost independently of each other and after properly linked by the deontic dimension. The organizational models that follow the organizational centered point of view (e.g., AALAADIN [5], MOISE [9]) usually are composed by two core notions: an Organizational Specification (OS) and an Organizational Entity (OE). An OE is a population of agents functioning under an OS. We can see an OE as an instance of an OS, i.e., agents playing roles defined in the OS (role instance), aggregated in groups instantiated from the OS groups, and behaving as normalized in the OS. Following this trend, a set of agents builds an OE by adopting an appropriate OS to easily achieve its purpose. An MOISE+ OS is formed by a Structural Specification (SS), a Functional Specification (FS), and a Deontic Specification (DS). Each of these specifications will be presented in the sequel.

Method. In MOISE+, as in MOISE, three main concepts, roles, role relations, and groups, are be used to build, respectively, the individual, social, and collective structural levels of an organization. Furthermore, the MOISE original structural dimension is enriched with concepts such as inheritance, compatibility, cardinality, and sub-groups. Individual level. The individual level is formed by the roles of the organization. A role means a set of constraints that an agent ought to follow when it accepts to enter a group playing that role. Following [2], these constraints are defined in two ways: in relation to other roles (in the collective structural level) and in a deontic relation to global plans (in the functional dimension). In order to simplify the specification1, like in object oriented (OO) terms, there is an inheritance relation among roles [6]. If a role ρ′ inherits a role ρ (denoted by ρ ⊏ρ′), with ρ ̸= ρ′, ρ′ receives some properties from ρ, and ρ′ is a sub-role, or specialization, of ρ. In the definition of the role properties presented in the sequence, it will be precisely stated what one specialized role inherits from another role. For example, in the soccer domain, the attacker role has many properties of the player role (ρplayer ⊏ρattacker). It is also possible to state that a role specializes more than one role, i.e., a role can receive properties from more than one role. The set of all roles are denoted by Rss.

Following this OO inspiration, we can define an abstract role as a role that can not be played by any agent. It has just a specification purpose. The set of all abstract roles is denotated by Rabs (Rabs ⊂Rss). There is also a special abstract role ρsoc where ∀(ρ∈Rss) ρsoc ⊏ρ, trough the transitivity of ⊏, all other roles are specializations of it.

Social level. While the inheritance relation does not have a direct effect on the agents behavior, there are other kinds of relations among roles that directly constrain the agents. Those relations are called links [9] and are represented by the predicate link(ρs, ρd, t) where ρs is the link source, ρd is the link destination, and t ∈{acq, com, aut} is the link type. In case the link type is acq (acquaintance), the agents playing the source role ρs are allowed to have a representation of the agents playing the destination role ρd (ρd agents, in short). In a communication link (t = com), the ρs agents are allowed to communicate with ρd agents. In a authority link (t = aut), the ρs agents are allowed to have authority on ρd agents, i.e., to control them. An authority link implies the existence of a communication link that implies the existence of an acquaintance link:

Collective level. The links constrain the agents after they have accepted to play a role. However we should constrain the roles that an agent is allowed to play depending on the roles this agent is currently playing. This compatibility constraint ρa ▷◁ρb states that the agents playing the role ρa are also allowed to play the role ρb (it is a reflexive and transitive relation). As an example, the team leader role is compatible with the back player role (ρleader ▷◁ρback). If it is not specified that two roles are compatible, by default they are not. Regarding the inheritance, this relation follows the rule Roles can only be played in the collective level, i.e., in a group already created in an OE. We will use the term “group” to mean the instantiated group in an OE and the term “group specification” to mean the group specified in an OS. Thus, a group must be created from a group specification represented by the tuple A group can have intra-group links Lintra and inter-group links Linter. The intra-group links state that an agent playing the link source role in a group gr is linked to all agents playing the destination role in the same group gr or in a gr sub-group. The inter-group links state that an agent playing the source role is linked to all agents playing the destination role despite the groups these agents belong to.

Conclusion. In this paper, we have presented a model for specifying a MAS organization along the structural and functional dimension, which are usually expressed separately in MAS organization models as we have stressed in the introduction. The main contribution of this model is the independence design of each one of these dimensions. Furthermore, it makes explicit the deontic relation which exists between them. We have used the MOISE+ model to properly specify the three dimensions of a MAS organization in both a soccer domain, used as an example here, and in a B2B (business to business) domain, not presented here. Comparing this proposal with the MOISE model [9], on which this work is based, the contributions in the structural dimension aim, on one hand, to facilitate the specification with the inclusion of an inheritance relation on the roles, and on the other hand, to verify if the structure is well formed, with the inclusion of the compatibility among roles and of a cardinality for roles and groups. Regarding the functional dimension, the main contributions are: the changes in the mission specification in order to express the relation among goals and their distribution through the inclusion of SCHs in the model; the inclusion of the preference among missions; and the inclusion of time in the deontic relations. Its functional specification is represented in a high abstraction level. Nevertheless, this specification could be specialized in a more detailed functional description already developed in the MAS area. For instance, a SCH could be detailed in a TÆMS task description [4] without redefining the structural specification. Even if an organization is useful for the achievement of a global purpose, as mentioned in the introduction, it can also make the MAS stiffer. Thus the system may loose one important property of the MAS approach, its flexibility. For example, if the environment changes, the current set of allowed organizational behaviors may not fit the global purpose anymore. In order to solve this problem, a reorganization process is mandatory.