2
Agent-based Modeling of Human Organizations

2.1. Introduction

This section has a narrow spine but a wide embrace. In addressing the relationship between agents and organizations, it takes in an extensive but highly fragmented set of ideas and studies embedded in organization theories. Its purpose is to try to devise a common ground model from which organization theories of various sorts can be logically derived.

Many theories have been developed to explain how organizations are structured and conducted and how the stakeholders involved behave. Each one takes a definite point of view without giving the opportunity to understand how they are possibly correlated with each other and whether a mapping of a sort between some of them can be figured out. On the contrary, founders of new approaches seem in most cases to ignore previous works. Agent ontology can function as a background model, allowing for the derivation of the main organizational theories from this base.

2.2. Concept of agenthood in the technical world

2.2.1. Some words about agents explained

In the technical field, the concept of agenthood is widely used. A general definition of what an agent is has been produced by J. Ferber (1999) (p. 9). Its adaptation is as follows:

Box 2.1.

An agent is a physical or virtual entity:

• a) that is capable of acting in an environment;
• b) that can communicate directly with other agents;
• c) that is driven by trends towards objectives of the sort of individual and/or collective satisfaction;
• d) that has access to resources for achieving its goals expressed in terms of objectives;
• e) that can perceive its environment commensurately with its objectives;
• f) that possesses skills and competencies for delivering services;
• g) that may be able to self-reorganize for survival in its current environment;
• h) that is endowed with autonomy.

All the terms in italics describe the key features of an agent (action, communication, objectives, autonomy and availability of resources). Virtual entities are software components and computing modules. They are not accessible to human senses as such but, as their number grows exponentially in our living ecosystem, they are contrived to become more and more the faceless partners of human beings.

According to their main missions, specific names have been given to agents, namely communicating agents (computer environment without perception of other agents), situated agents (perception of the environment, deliberation on what should be done and action in this environment), reactive agents (drive-based reflexes) and cognitive agents (capable of anticipating events and preparing for them).

It is important to contrast object, actor and agent. In the field of computer science, object and actor are conceived as structured entities bound to execute computing mechanisms. An object is depicted by three characteristics:

• - the class/instance relationship representing the class as a structural and behavioral meta-model and the instance as a concrete model of the context attributes under consideration;
• - inheritance enabling one class to be derived from another and benefiting from the former in terms of attributes and procedures;
• - message discrimination to trigger polymorph procedures (methods in data-processing vernacular) as a function of incoming message contents.

The delineation between objects and communication agents is not always straightforward. This is the fate of all classifications. If a communication agent can be considered as an upgraded sort of object, conversely an object can be viewed as a degenerate communication agent whose language of expression is limited to the key words corresponding to its methods.

An agent has services (skills) and objectives embedded in its structure, whereas an object has encapsulated methods (procedures) triggered by incoming messages.

Actors in computer science perform data processing in parallel, communicate by buffered asynchronous messaging and generally ask the recipient of a message to forward the processed output to another actor.

Another concept associated with agents is what is called a multiagent system (MAS) (Ferber 1999). It has become a paradigm to address complex problems. There is no unified, generally accepted definition but communities of practice. The approaches followed by the system designer, namely functional design or object design, are chosen on the basis of answering the two following questions:

• - What should be considered as an agent to address the issues raised by the problem to tackle?

A system is analyzed with a functional approach when centered on the functions the system has to fulfil or with an object approach when centered on the individual or the product to deliver.

• - How are the tasks to perform allocated to each agent in the whole system from a methodological point of view?

There is no miracle recipe to achieve a good design. In addition, it is possible to analyze the same system from different angles and to deliver different designs. The approach for coming to terms with a problem is influenced by the historical development of the field involved. Many people are inclined to think of MAS as a natural extension of design by objects.

It is worth noting that the MAS paradigm has disseminated in many technical areas where centralized control was a common practice. For many a reason, especially the computing power of on-board systems and the reliability of available telecommunication services, coordination between distributed systems takes place directly between the very units of the system without any central controlling device. This situation already prevails in railway networks.

2.2.2. Some implementations of the agenthood paradigm

The concept of agenthood has been applied in various technical fields from the 1990s onwards. Two examples will be described here, namely telecommunication networks and manufacturing scheduling.

2.2.2.1. Telecommunications networks

The world of telecommunication networks is extensively modeled on the basis of this concept. A telecommunication network is a mesh of nodes fulfilling a variety of tasks. Each node is an agent. It can be defined as a computational entity:

*acting on behalf of other entities in an autonomous fashion (proxy agent);

*performing its actions with some level of proactivity and/or reactiveness;

*exhibiting some level of the key attributes of learning, cooperation and mobility.

Several agent technologies are operated mainly in the telecommunications realm. They fall into two main categories, i.e. distributed agent technology and mobile agent technology.

Distributed agent technology refers to a multi-agent system described as a network of actants with the following advantages:

• - solving problems that may be too large for a centralized agent;
• - providing enhanced speed and reliability;
• - tolerating uncertain data and knowledge.

They include the following salient features:

• - communicating between themselves;
• - coordinating their activities;
• - negotiating their conflicts.

"Actants" are non-human entities such as configurations of equipment, mediators and software programs and are distinguished from actors that are human beings. But actors and "actants" are entangled in ways that provoke complexity dynamics in many circumstances.

Mobile agent technology functions by encapsulating the interaction capabilities of agents into their descriptive attributes. A mobile agent is a software entity existing in a distributed software environment. The primary task of this environment is to provide the means which allow mobile agents to execute. A mobile agent is a program that chooses to migrate from machine to machine in a heterogeneous network.

The description of a mobile agent must contain all of the following models:

An agent model (autonomy, learning, cooperation).

A life-cycle model: this model defines the dynamics of operations in terms of different execution states and events, triggering the movement from one state to another (start state, running state and death state).

A computational model: this model, being closely related to the life-cycle model, describes how the execution of specified instructions occurs when the agent is in a running state (computational capabilities). Implementers of an agent gain access to other models of this agent through the computational model, the structure of which affects all other models.

A security model: mobile agent security can be split into two broad areas, i.e. protection of hosts from malicious agents and protection of agents from hosts (leakage, tampering, resource stealing and vandalism).

A communication model: communication is used when accessing services outside of the mobile agent during cooperation and coordination. A protocol is an implementation of a communication model.

A navigation model: this model concerns itself with all aspects of agent mobility from the discovery and resolution of destination hosts to the manner in which a mobile agent is transported (transportation...