International Encyclopedia of Systems and Cybernetics

A technique leading to "… the construction of so-called multi-agents systems which permit modelization of the behavior of a set of more or less expert entities, more or less organized along kinds of social laws" (J. ERCEAU & J. FERBER, 1991, p.751).

ERCEAU and FERBER explain that "These entities or agents are endowed with a measure of autonomy and are immersed in an environment wherein and with which they interact. Accordingly, their structure is organized along three basic functions: to perceive, to decide and to act. They are physical entities (sensors, processors, vehicles…) or abstract ones (tasks to be executed, displacements,…), able to act upon their environment and upon themselves, i.e. to modify their own behavior. To do this they are endowed with a sketchy representation of that environment, and with perceptual and communication abilities" (Ibid).

The striking likeness of D.A.I.'s basic concepts with the basic working conditions of human societies may well be a harbinger of the shaping of societies of cooperative automata.

According to the authors the characteristics of these automata are of two different kinds. They should include:

- "eco-behavior" programs as for example satisfaction of needs, escape, aggression, return to satisfaction which would allow them to "… better their individual and community performance through learning-like mechanisms".

- specialized modules within the structure of every agent. We should thus have "communicating agents, rational ones, intentional ones and reactive ones" (Ibid).

Structure of the agents

J. ERCEAU and J. FERBER distinguish and characterize in the following way the agents in D.A.I.:

Reactive agents: "They are of the lowest class. They only have at their disposal a reduced protocol and communication language and… their capabilities correspond merely to a stimulus-action law. The class of reactive agents include various levels, according to their capacity to gather and produce global behavior".

Communicating agents: "… which have at their disposal a complete communication protocol, but whose conversational and behavioral sections are mutually dependent… The agents, at that level are unable to process such informations which are off the limits of their expertise domain" and thus "… the exchanges of informations are closely tied to the expertise existing within their domain".

Relational agents: "Each agent has at his disposal precise abilities, beliefs and a partial representation of its environment, including the other agents in the system. Task allocations mechanisms are incorporated for specific practical applications".

Intentional agents: They constitute the upper level: "Each of them is endowed with specific goals, partial plans allowing it to reach its goals, with intentions and beliefs, and the possibility to contract commitments which it must carry out, or contract other agents to perform some actions" (1991, p. 757-8).