As already hinted, a main reason for the shift from SAS to MAS can be identified in the inescapable need of effectively integrating, in a scalable way, people and computer systems within real, complex workplaces. Socio-economic and psychological literatures view them as instances of human organizations, for which they worked out terminologies and models widely borrowed (and, indeed, also tested and in some cases enhanced through formalizations, simulations and implementations) by researchers in Distributed Artificial Intelligence [Gasser, 1991; Werner, 1992], the area that gave origin to the MAS field.
What characterizes an organization? According to [Conte and Castelfranchi, 1995], groups or organizations can be viewed as social systems endowed with macro-level goals influencing individual agent's behaviors so that they are or become adapted to pursue those purposes. Social actions can occur only in the presence of an often complex network of conditions and relations that account for the social system structure. More specifically, for the social system to be adequately "implemented", the member agents must be bound each other through social commitments [Gasser, 1991], that is relations expressing the engagement of an agent to perform some action in behalf of (at least) another one [Castelfranchi, 1995a]. This means that an organization is "more" than the sum of the properties and behaviors of its members, which still enjoy a "socially bounded autonomy" allowing them to pursue individual (micro-level) goals at least through self-sufficient task execution (executive autonomy [Castelfranchi, 1995b]). But for an organization to show reliable and stable courses of activity some more inter-agent relationships are needed, expressing the roles of mutual dependence and power/authority (e.g. within a hierarchy), that account for a shared "awareness" of the need to cooperate. For example, such relations, determining the organization structure along with its normative content and thus its level of stability and activity coherence, can be expressed as meta-commitments, that is, commitments to commit [Castelfranchi, 1995a]. These structuring links, known by member agents, distinguish organizations, groups and teams from other forms of "extemporaneous" social systems or communities. Sharing the same goal, for example, does not represent a sufficient condition for constituting an organization, as the agents in question may hardly compete with each other e.g. to get the needed resources.
The need of formalizations expressing the underlying framework of composite organizations exploiting computer-based systems was captured also by other projects, like those converging into the Enterprise Project, aiming at providing a methodology and a toolset for enterprise modelling, thanks to the specification of structural concepts in a central Enterprise Ontology [Uschold et al., 1995]. This extensive and quite complex ontology, written in the standard language Ontolingua [Gruber, 1993], includes as a basic component a theory of organizational structure general enough for adapting also to non-business frameworks: for example, it defines concepts like ORGANIZATIONAL UNIT, PERSON and MACHINE. The same could be said for other subtheories specifying concepts inspired from the area of multi-agent (therein called actors) planning, like those of activity, plan and resource. The main role taken by the ontology is that of a mediating mechanism between people alone, people and computational systems, computational systems alone and finally enterprises as a whole. The general model is intended to be further specialized into a set of models to be stored in a library, as advocated also e.g. in [Dieng et al., 1995].
Moreover, the need of a "global semantics"-based social foundation in order to address well-known MASs issues like the joint qualification and the representational incommensurability problems, arising respectively from the impossibility, for an agent, of fully specifying the assumptions and the semantics of its local knowledge to another agent, so that it becomes undecidable whether the two can coherently cooperate or not (i.e. they are in conflict), was recognized by DAI founding fathers like Les Gasser [Gasser, 1991].
Within such organizations, various types of interactions may become feasible, being function of the structural relationships among agents (roles, dependence). In a strongly normative institution, there will be limited space for task bargaining among agents. For example, in [Conte and Castelfranchi, 1995] an orchestrated cooperation model is devised, in which only the "boss agent" knows and enforces the cooperative plan, assigning roles and thus goals and tasks to other agents. We call this kind of interaction, in which a dominant legitimate authority exists, coordination. Coordination limits the amount of necessary inter-agent communication, tries to eliminate situations of conflict and avoids antagonistic or non cooperative behaviors, like those illustrated by [Werner, 1992] in his diagram representing the so-called "Spectrum of Cooperation Styles". In the case of a D-HIS for a hospital, the enforcement of official guidelines, indicating treatment procedures for given classes of diagnosed diseases, can be suitably obtained creating an agent possessing the global procedure (the guideline manager) and delegating steps or single tasks of the latter to other ones according to their institutional competence.
When instead a substantial margin is left to tasks and services negotiation among agents, we will speak of "proper" cooperation. Cooperation requires, in general, a richer ACL (that is, more or more structured agent communication primitives), may be itself object of negotiation, and may depend on the problem at hand. For example, it may be established in an occasional, temporary or permanent way. In the same D-HIS, health care providers may set such cooperation links with each other whenever in an occasional, short-term or long-term perspective they need to consult another specialist or to benefit from a particular service they are not directly competent or authorized for.