artificial intelligence: agent technology
DESCRIPTION
This presentation covers agent technology for artificial intelligence. Topics covered are as follows: expert systems, overcoming expert systems limitations, agent, what is an agent, definition of an agent, agents versus expert systems, how is an agent different from other software, types of agents, deliberate versus reactive, interface versus information, mobile versus stationary, and why a mobile agent.TRANSCRIPT
Agent Technology
• Expert System (ES) vs Agent
• Definition
• Types of Agents
Expert Systems• Rule Based
• Case Based
• Knowledge system separate from inference engine
• Pros De-coupling of the system knowledge from the
inference mechanism supports easier maintenance and management of both
System architecture can be readily re-used for different applications by, for instance, using different FACTS and RULES
Financial/Engineering/IS
Expert Systems• Cons/Limitations
– Information Brittleness: If the environmental input is outside of the expert system's fact and/or rule-base scope it does nothing (quits)
• Set of Rules: User wants something outside of fixed scope
• Exception handling: Out of Scope
– Isolation: An expert system is stand-alone i.e., does not enter into "collaboration with other expert systems”
• Inefficient: Interrelated but no communication across ES systems
• Not enterprise level
– Example: ES 1 Finance/ES 2 Accounting
– Static Behavior: The Level of Behavior (LOB) of the expert system is static, i.e., it does not improve over time or use
– Incapable of Learning
• Should be able to learn from mistakes & improve
How to overcome the ES limitations• How to overcome brittleness:
– Support graceful degradation of performance by initiating and maintaining dialog with the environment (users)
• Instead of just stopping, try to increase scope gradually by including user into system
– Expand fact and rule bases automatically through learning mechanisms
• Introduce new variable, ok first time, next time will recognize
• How to overcome isolation:
– Add mechanisms for coordination and collaboration
• ES 1 Finance/ES 2 Accounting
– Permit communication with each other at enterprise level
How to overcome the ES limitations
• Static:
– provide learning mechanisms
Agent
The augmentation of expert system capabilities to overcome limitations leads, in an evolutionary fashion, to the creation of intelligent agents Sharing knowledge
Increase scope
What is an Agent
• A computer system that – has goals, perceptors, and effectors– decides autonomously which action to take
in a given situation
What is an Agent• Use networking to share data across enterprise• No longer independent• Becomes an “Agent Community” also includes “Human User”• Resolves isolation, brittleness, static
ES 1Finance
Fixed Scope
ES 3
Fixed Scope
ES 2Accounting
Fixed Scope
ES 4
Fixed Scope
Hub
OOP Similarity
• OOP Class uses Method invocation• Agent happens dynamically• OOP
– (Method invoked) + (Rich Symantec/syntax) = Agent– OOP (not intelligent) + ES (intelligent) = Agent– Object just performs function– 1) Agent can calculate or can reject (can think/then perform)– 2) Communication ability
Methodology - Action
Intelligent
Methodology - Action
Not intelligent
Class Def - Public/PrivateClass Def - Public/Private
Methodology
Class Def
The Class is the Intelligent Object
The OOP resides on top of the ES and creates an AI Agent
Definition of an Agent• Autonomy = Intelligent
– Can reason/can think/learns from previous experience and can apply to new experience
– an agent operates without the direct intervention of humans or others and has control over its own actions
– an agent is able to exhibit goal-directed behavior by taking initiatives
– Knows what problem to solve (goal directed behavior)• Social ability
– an agent interacts with other agents via Agent Communication Language (ACL)
• Symantec/Syntax of Language• Adaptive
– an agent learns to improve its behavior
Agents vs ES
• Personalized– agents -> different actions– ES -> same actions
• active, autonomous– agents - on their own– ES - passively answer
• adaptive– agent - learn and change– ES - remain static
How is Agent different from other Software
• Personalized, customized
• Proactive, takes initiatives
• Autonomous
• Adaptive
Types of Agents
Agents
Architecture Functionality Mobility
Deliberative Reactive Interface Information/ Mobile StationaryAgent Agent Agent Internet Agent Agent
Agent
Deliberate vs Reactive
• A deliberative agent has an internal representation of the sequence of actions necessary to achieve a goal given or an event triggered (a Pre-plan/Priori plan).
• Reactive agents do not store a priori plan of the actions. No internal representation of pre-plan exists within any of the reactive agents and, hence, the plans has to emerge upon an event through collaboration.
Interface vs Information
• Interface agents migrate from the direct command metaphor to one that delegates some of the tasks to the agents in order to accommodate novice users.
• Information (Internet) agents are designed to deal with the problem of information overload and the general issues of information management in Internet.
Mobile vs Stationary
• Mobile agents are able to roam the network such as WWW, interacting with foreign hosts, and performing the various duties assigned at a remote site
• Stationary agents stay at the client or at the server.
Why Mobile Agent?
• Reduces network traffic
• Shares load among machines
• Go to the data if the data can’t come to you
• User may have only infrequent connection to network