Distributed Artificial Intelligence
Research Program 2000-2003
Problems
Our objectives
led us to address a number of problems related to the methods of analysis
and design of systems of cognitive agents, as well as to the development
of a generic model of agent, and to the various concepts and mechanisms
necessary to develop such systems.
Methods
of analysis and design for multi-agent systems
One of the major difficulties when trying to develop systems of cognitive
agents is encountered during the phase of analysis and design. In some
applications, it is very hard to determine the number and the nature of
the agents to be implemented, and to select the needed communication and
negociation protocols.
Model of a generic
agent
The design of multi-agent systems would benefit from the existence
of a generic model of agent. However, currently no such model is available.
Architecture
of systems of cognitive agents
The architecture of such systems is an extremely difficult problem
which requires to address a number of different issues. The principal issues
are the following:
the concept of open systems
Complex systems must allow dynamic changes without
perturbing the overall functioning. I.e., one must be able to add, suppress,
or modify agents, without halting the system. The concept is in itself
very simple, its implementation is very difficult.
multi-level architecture
In some cases, for distributed applications, one can envision to build
groups of local agents , and distinguish between
local interactions and interactions among agents belonging to different
groups.
integration of human actors
In most applications of interest to us, the human element cannot be ignored.
Our systems of agents are thus mixed and contain automatic agents and human
agents. This led us to propose a proxy agent,
a specific agent used as an interface between a human
actor and the rest of the system.
transfer of knowledge among agents
When agents exchange not only data but also knowledge, one need to specify
the concepts that are used during such exchanges. Modeling concepts belongs
to the area of research dealing with ontologies.
design of development environments
The development of actual multi-agent platforms involves
many problems, like that of communication protocols, negociation protocols,
distribution among various machines. In addition one requires tools
which are currently not available.
Last modified 8/4/02