University of Msila

FACULTY OF MATHEMATICS AND

INFORMATICS
DEPARTMENT OF COMPUTER
SCIENCE

Introduction to DAI and Muti

Agent Systems
(MAS)
Outline
 Introduction
 DAI
 Agent concept
 Multi-agent systems
Introduction
 A.I. is the science and engineering of building intelligent machines with calculable,
intelligent, like human-behaviors:
 A machine can : Play chess, Talk, Translate text, Ride a bike, Bring breakfast to bed,
Recognize a friend in a photo.
 1970: Creation of Distributed Artificial Intelligence
 A new AI: Distributed AI

Objective:

 Create a society of autonomous agents working together towards a global goal

3
Introduction

Distributed solving Solving of distributed Resolution by

of problems problems coordination

Distribute the
AI

D.A. I
4
DAI
 Why distribute the A.I.?
The reasons for distributing artificial intelligence can be found in the following investigations:
 Simplifying distributed IT applications
 Distributed solving of problems

- A complex task is solved by a group of specialists with complementary skills,

- Expertise is distributed, but the domain is not,

- Ex: medical diagnostics, product design, pattern recognition...

 Solving of Distributed problem

- The domain is distributed,

- Analysis, identification and control of physically distributed systems,

- Ex: communication network control, road network control.

 Coordinated problem solving

- Expertise through coordination,

- Ex: assigning tasks in a workshop.

- Overcoming user interface problems,

DAI
 Example:
Functional distribution in human activities (such as product design, for example) :
Decomposition of the problem according to specialties
 DAI
 Parallel Artificial Intelligence
•Language and algorithm development for DAI
•Improving the performance of DAI systems by proposing concurrent languages
and parallel architectures

 Distributed Solving of Problems

•Decomposition of a problem posed on a set of distributed, cooperating entities
•Knowledge sharing between entities
•Entities are generally dependent on one another.

 Multi-Agent Systems
Cooperate a set of proactive and relatively independent entities called "agents",
endowed with intelligent behavior. with the aim of coordinating their goals and
action plans to solve problems.
Concept of agent
 "An agent is a mechanical, biological or software system that interacts with its
environment. Anne Nicole.
 For example:
 a printer can be seen as a mechanical agent that reacts to commands and produces
actions in return.
 Plants, animals and humans are biological agents with greater autonomy, absorbing
nutrients, breathing, transforming themselves and their environment.
 In Computer science: Software agents are autonomous programs, run on a machine,
which perceive certain elements of their environment via input streams (keyboard,
mouse, sensors) and act via their output streams (screen displays, physical machine
control, process control).
 Concept of agent
 Definitions
 An agent is an entity (physical or abstract) characterized by :
- Its autonomy in decision-making,
- Its knowledge of itself and of others,
- Its ability to act. J. Ferber and G. Ghallab, 88
 An agent: an intelligent entity, acting rationally and intentionally, according
to according to its own goals and the current state of its knowledge.
Y. Demazeau & J.P. Müller, 90

Two dimensions
J. Erceau & J. Ferber

Individual Social
9
 Concept of agent
 Behavior
Its behavior tends to satisfy its objectives, taking into account the resources and skills at its disposal, and
depending on its perceptions, representations and the communications it receives.
 Knowledge
An agent has three types of knowledge:
- domain knowledge or expertise ;
- control knowledge or behavior ;
- communication and interaction knowledge
 Agent characteristics
 Nature: An agent can be a physical or virtual entity.
 Autonomy: An agent is more or less independent of the user, other agents and resources (CPU,
memory, etc....).
 Environment: this is the space in which an agent will act; it can be reduced to the network made up of
all the agents.
 The objective: an agent can pursue the overall system goal, satisfy its own objectives or even behave
in a way that absolves itself of a survival function.
 Perception: of the environment by an agent.
 Communication: an agent has the ability to communicate with other agents.
 Reasoning: an agent may be linked to an expert system or other more or less complex reasoning
mechanisms.
Agent versus Object
 Autonomy: agents have control over their actions, they can refuse to cooperate
 Agents are reactive, like objects, but also proactive.
 Agents are usually persistent, and have their own thread of control.
Concept of agent

 Agent architecture

Agent

Contrôle Knowledge

Perception Communication

Environnement
Environment Agent community

control flow
: Data flow
Concept of agent
 Agent knowledge
 Domain knowledge
 Control knowledge
 Intentions
 Beliefs
 Decisions
 Rationalities
 Commitments

 Communication knowledge
 Communication expertise
 Messages
Concept of agent
 Agent types
There are two types of agent: reactive and cognitive.
 Cognitive agent
 Cognitive agents have a symbolic representation of their environment and reasoning
capabilities.
 Agents are immersed in an environment in which they interact. Hence their structure is
built around three main functions: perceiving, deciding and acting.
 We can also mention some important sub-functions:
- conflict detection,
- belief revision,
- cooperation (negotiation, coordination),
- learning.
Concept of agent
 How a cognitive agent works

Perception Beliefs Planning

other Decision
Agents communication goals
Expertise

Plans
World
Outdoor
exécution Intentions
Concept of agent
 Reactive Agent
 It does not include reasoning; it acts according to a stimulus/reaction pattern to
events produced as inputs.
 The "reactive" school, on the other hand, claims that it is not necessary for
agents to be individually intelligent for the system to behave intelligently as a
whole.

Example
rules : règles condition-action
percepts : ensemble de percepts
repeat
state := interpret_input(percept) ;
rule := match(state, rules) ;
execute(rule[action]) ;
forever
Concept of agent
 Difference between a Reactive Agent and a Cognitive Agent

Cognitive agent system Reactive agent system

Explicit representation of the environment No explicit representation

Can take its past into account No history memory

Complex agent Stimulus/reaction operation

Small number of agents Large number of agents