Agents and Multiagent Systems Chapter 6 Dr Ahmed Rafea
Transition from AI to IA • There are many alternative artificial intelligence techniques for knowledge representation , reasoning and learning. • The specific functions and requirements of an intelligent agent are the prime determinant of which AI technique should be used.
Knowledge representation • Knowledge representation is a crucial issue. • What our agent is expected to do and in what domain, will have a significant impact on the type of knowledge representation we should use.
Reasoning • The amount of intelligence required by an agent , in terms of the size of the knowledge base and sophistication of the reasoning algorithms , is significantly impacted by the degree of autonomy and mobility the agent has . • Mobile agents place special requirements on the security of the knowledge base it travels through the network.
Learning • Whether learning is a desirable function depends on the domain the intelligent agent will work in , as well as the environment. • Learning is most useful when an agent is used in complex environments to perform repetitive tasks , or when the agent must adapt to unknown situations.
Autonomous Intelligent Agents Requirements for Autonomous Intelligent Agents include: • Perception • Taking Action
Perception • In order for a software agent to take some intelligent action , it first has to be able to perceive what is going on around it. • An intelligent agent uses its sensors as a source of information. • A fundamental part of perception is the ability to recognize and filter out the expected events and attend to the unexpected ones.
Taking Action • Intelligent Agents use effectors to take actions either by sending messages to other agents or by calling application programming interfaces or system services directly. • If our agent takes an action directly under its control, we can probably consider it done. However, when we are dealing with intermediaries , whether other agent or unknown systems, then some extra precautions and checking are probably in order.
Multiagent Systems • Multiagent systems are applications in which many autonomous software agents are combined together to solve large problems. • The RoboCup challenge is an example of the current state-of-the-art of multivalent systems , in which teams of autonomous agents compete in a simulated soccer tournament.
Blackboards • Blackboard is the oldest multiagent system architecture used as a problem-solving technique. • The Blackboard is a data structure that is used as the general communication mechanism for the multiple knowledge sources and is managed and arbitrated by a controller .
Blackboards • As each agent works on its part of the problem, it looks to the blackboard to pick up new information posted by other agents , and it, in turn, posts its results to the Blackboard. • Blackboard systems are used as a communication mechanism when building single large applications and want to modularize the knowledge bases.
Communication • An environment where agents with very different structures and with no knowledge of a centralized background can work together, the agents will need to communicate with each other. • Communication can be : • Directly to each other • Through an interpreter or facilitator.
Communication • To be able to communicate, a language is needed. • There is a level of basic language which is the syntax and format of the messages and there is a deeper level, the meaning or semantics . • For the semantics to be easily understood , a shared vocabulary of words and their meanings is needed. This shared vocabulary is called an ontology . • The most widely used agent communication language (ACL) is Knowledge Query and Manipulation Language (KQML).
Knowledge Query and Manipulation Language • Knowledge Query and Manipulation Language (KQML) provides a framework for a set of independent agents to communicate and cooperate on a problem using messages called per formatives . – Directives: commands or requests – Representatives: facts or beliefs – Commissives: promises or threats • KQML uses ontologies to ensure that two agents are communicating in the same language • KQML messages encode information at three different architectural levels: content, message and communication. An example of a KQML message from agent joe asking about the price of a share of SUN stock might be encoded as: (ask-one :sender joe --comm. level :content (real price = sun.price()) --content level :receiver stock-server --comm. level :reply-with sun-stock --comm. level :language java --message level :ontology NYSE-TICKS) --message level
Agent Standards • Standards are becoming more important as agents become a large part of the electronic commerce infrastructure. • Two major efforts of standardization are: – The Foundation for Intelligent physical Agents (FIPA) that is focused primarily on agent-level issues. – The Object Management Group (OMG) that is focused on object-level interoperability and management
FIPA & OMG • FIPA is dominated by computer and telecommunications companies and is focused primarily on agent-level issues. • OMG is the standards body that created the C ommon O bject R equest B roker A rchitecture (CORBA) distributed object protocol and tends to focus on object-level interoperability and management.
Co-operating Agents • Co-operation among agents allows a community of specialized agents to pool their capabilities to solve large problems but with the additional cost of communication overhead. • Distributed systems management, electronic commerce and multi agent design systems are three application areas in which co-operating agents have been applied. • It is likely that a combination using the team structure and roles to limit communications , along with distributed planning techniques , will provide the best solution to building multiagent teams.
Competing Agents • Competition between agents will occur as soon as intelligent agents are deployed by individuals or companies with different agendas and those agents interact in the e-commerce environment • Intelligent agents will be used to provide advantages for individuals and businesses. • Negotiation protocols, such as Contract Net , auctions and bargaining, are used to allow agents to compete for business.
Agent Software Engineering Issues • Designing multiagent systems is similar to object-oriented but requires some additional analysis and modeling techniques. • A common approach for designing agents and multiagent systems is to define roles for team members. • While agent applications are becoming increasingly popular, there have not been many proposals for agent-oriented methodologies for analysis, design, and software development.
Designing Agents Two Popular methods are: • The agent modeling technique for systems of agents. This approach looks to the problem from two perspectives: an external and internal one. – External: The agents themselves (agent Model) and their interactions (interaction model) – Internal: Relationship with other agents, a goal, and a plan to achieve the goal • CoMoMas extension to the CommonKADS knowledge engineering methodology.
Multi Agent Interaction Models • Model 1: 2-Request 1-Request Requesting Service Agent Facilitator Agent 4-Response 3-Response
Multi Agent Interaction Models • Model 2 1-Request 2-Request Requesting Service Agent Facilitator Agent 3-Response
Multi Agent Interaction Models 3-Request • Model 3 1-Request Requesting Service Agent Facilitator Agent 2- Address of service Agent 4-Response
Recommend
More recommend
