Multi-Agent Systems


The aim of multi-agent systems is to understand how independent processes can be coordinated.

An agent is a computerized entity like a computer programme or a robot. An agent can be described as autonomous because it has the capacity to adapt when its environment changes. A multi-agent system is made up of a set of computer processes that occur at the same time, i.e. several agents that exist at the same time, share common resources and communicate with each other. The key issue in multi-agent systems is to formalize the coordination between agents. Research on agents, therefore, includes research into:

  1. Decision-making : what decision-making mechanisms are available to the agent? What is the link between their perceptions, representations, and actions ?
  2. Control : what hierarchic relationships exist between agents ? How are they synchronized ?
  3. Communication : what kind of messages do they send each other? What syntax do these messages obey ?

Multi-agent systems can be applied to artificial intelligence. They simplify problem-solving by dividing the necessary knowledge into subunits-to which an independent intelligent agent is associated-and by coordinating the agents' activity. In this way, we refer to distributed artificial intelligence.

This method can be used for monitoring an industrial process, for example, when the sensible solution -that of coordinating several specialized monitors rather than a single omniscient one- is adopted.

Fundamental research is being conducted on the problems associated with the representation of agents' decisions and protocols for communication. The main applications for MAS are for telecommunications, internet, and physical agents, such as robots. A group of scientists has specialized in the simulation of agents' societies in the fields of ecology and social sciences. Research on modelling distributed systems has also been conducted by computer scientists, ecologists, and social scientists. Here, we discuss a few references and some of the issues involved.

MAS and Ecology

In ecology, the distributed approaches or individual-based models (IBM) were developed at the end of the 1980s. The most frequently quoted article is that written by Huston who puts forward two reasons for developing this approach: firstly, there is the need to take into account the individual because of their genetic uniqueness and, secondly, the fact that each individual is situated and their interactions are local. The argument clearly had an impact because numerous publications refer to this approach. Shortly after Huston's publication, Hogeweg and Hesper published a similar article on "individual-oriented modelling". The latter had more influence on scientists working in the field of artificial life. Then, in 1990, there was a congress in Knoxville on "Individual-based models and approaches in ecology". The congress proceedings were published and have since been the main reference text on the subject. Above all, truly individual-based models, or so-called i-state configuration models have been used for the object-oriented approach, for example, the work by Silvert, Derry, and Roese. There are several models of forests. The most striking application is undoubtedly the Across Trophic Level System Simulation (ATLSS) model which attempts to simulate the ecological function of the Everglades in Florida. This model represents abiotic factors-such as hydrology, fire, and hurricanes-and different trophic levels simultaneously. Within the models, which can be mathematical compartmental models, different animal populations are simulated (deer, fish) with the help of individual-based models. Craig Reynolds proposes a rich list of classifyied and annotated links relative to IBM's.

MAS and social sciences

MAS are developing rapidly in the field of social sciences. Society simulation is the subject of numerous conferences, for example, Multi-agent Systems and Agent-Based Simulation (MABS). Research on the subject is published in the electronic journal Jasss (Journal of artificial societies and simulation). In addition, a group called Agent-based Computational Economics (ACE) has been set up.

In social sciences, the application of multiagent universes to simulate social phenomena is generally associated with the sociological trend, so-called "individualism", in which the singular individual is considered as the elementary unit or the atom of society. The overlap is, in fact, in the bottom-up approach which characterizes MAS. However, the assimilation between individuals from a society and agents from a multi-agent universe can be misleading. In fact, it is quite possible for social groups and institutions to be considered as agents with their own standards and rules. The agents are directed by constraints or rules that are expressed on a group level, i.e. they are no more than entities that act and are placed in a dynamic environment.

This straightforward comment -which is natural when MAS are used for modelling- shows how the simple duality that exists between individualism and holism can be called into question. It is a major preoccupation for researchers working on renewable resource management and MAS :

  1. individuals, products of history, are driven by collective values and rules,
  2. collective values and rules evolve because of the interaction between individuals and between groups,
  3. the individuals are neither similar nor equal but have their own specific roles and social status.

How do individuals make up a group ? How is an institution created ? The individual cannot be considered as an autonomous entity that is independent of its social environment. How are individuals constrained by the collective structures that they themselves have set up and how do they make these structures evolve ? What degrees of freedom are given to the definition of individual practices ? Here are just some of the questions that can be explored using MAS and which can be expressed as follows: "how are collective structures set up and how do they function when they are based on agents with different capacities of representation, that exchange information, goods or services, etc., draw up contracts, and are thrust into a dynamic environment which responds to their actions ?"

MAS and interactions between societies and resources

The problem with modelling common resource management is how to simulate the interaction between groups of agents and resource dynamics. There are several empirical ways of modelling these interactions.

The first method involves simulating how social networks are managed. Here, we consider that the relationships between people and resources should, in fact, be rephrased as the relationships between people who affect resources. Agents that exchange messages within networks or so-called contact networks can be simulated with MAS. In this way, it is possible to simulate exchanges of information and services as well as contracts and agreements between agents. For example, in the case of irrigated systems, farmer agents can send each other messages so that they know what the water levels are in different plots. They can also ask for or exchange services or addresses. In this way, it has been shown that the evolution of a system can be very sensitive to the structure and dynamics of social networks.

In the second method, emphasis is put on the cognitive processes or representations that determine how agents and resources interact. Each agent develops and then acts on its own representation of a resource. In so doing, the agent transforms the resource for the other agents. This model represents interactions that are close to what economists call externalities. We study the problem of managing common renewable resources by examining the different representations of actions that affect the resources in question. The resulting resource use may or may not be satisfactory for the agents. This can be described as coordination through the environment.

This method of classification provides a framework in which applied models can take into account both the agents' interactions and systems of exchanges through the environment.


Abott C.. A., M. W. Berry, et al.,1995. Computational models of white tailed deer in the Florida everglades, UTK-CS.

DeAngelis, D. L. and L. J. e. Gross,1992. Individual-based models and approaches in ecology, Chapman and Hall.

Deutschman, D. H., S. Levin, C. Devine, L. Buttel,1997. Scaling from trees to forests : analysis of a complex simulation model. (abstract)

Ferber, J.,1999. Multi-Agent Systems : an Introduction to Distributed Artificial Intelligence. Reading, MA, Addison-Wesley.

Gilbert, N.,1995.. Emergence in social simulation. Artificial societies. The computer simulation of social life. R. c. a. N. Gilbert, UCL Press: 144-156.

Hogeweg, P. and B. Hesper,1990. Individual-oriented modelling in ecology. Mathl. Comput. Modelling 13: 83-90.

Huston, M., D. DeAngelis, et al.,1988. New computer models unify ecological theory. Bioscience 38: 682-691.

Weiss, G., Ed. 1999. Multiagent Systems : a Modern Approach to Distributed Artificial Intelligence, MIT Press.


This page have been translated into the Belorussian language thanks to Martha Ruszkowski


Le Cirad Centre de coopération internationale en recherche agronomique pour le développement
Informations légales © Copyright Cirad 2001-2015